ConstructINS

What's new

News & updates

The latest news and updates for ConstructINS.

Release notes

Everything added to ConstructINS

Every update to the platform, newest first — 97 releases since 19 Sept 2026.

  • 166 new features
  • 123 improvements
  • 93 fixes
v1.96Latest

The homepage now answers what buyers asked first: who feeds it, what the AI sends outside Egypt, how it connects, and what the pilot figures are — with a page for the owner’s side and a technical sheet.

  • New“Nothing new to type on site”, right under the film: say the day’s work in English or Egyptian Arabic, photograph the delivery note, keep the WhatsApp Business number and the Excel — and no Primavera or ERP is needed to start.
  • NewA technical sheet at /spec, ready to print: every connection and how it works, the files the Upload center reads, sign-in (Microsoft Entra single sign-on and passkeys), hosting, records, export and the AI’s data handling.
  • NewRole views for the owner’s side and the outside parties — owner / developer, owner’s PM / PMO, consultant / resident engineer, public client and subcontractor — and three new product screens: the developer’s view, notice deadlines and the BOQ checker.
  • NewThe homepage headline is now an A/B test: half of visitors see the question “Which of your projects is about to slip — and what will it cost you?”. Views and demo requests per version are in the admin Forms inbox.
  • ImprovedWhat the AI sees, and where it goes, said plainly on the homepage, /trust and the FAQ: only the records your account may see; Claude, ChatGPT and Gemini may process it outside Egypt; your own Azure OpenAI keeps it in your region; and what still works with every provider off.
  • ImprovedIntegrations are listed as they really connect, from the connections catalogue — directly, or through file exports and webhooks — and Primavera P6 stays the master schedule: read, never written back.
  • ImprovedThe pilot section leads with what a buyer can check — the signed letter, the case study, how the figures were measured and by whom — and labels the four figures “as reported, not independently audited”. The pilot now starts from P6 or Excel.
  • ImprovedEach kind of project on the homepage has a worked example of its own: roads, utilities under Law 182, residential compounds, developer portfolios and PMOs — not only Tower A.
v1.95

The ConstructINS film is on the homepage: why construction projects run late and over budget, and what changes — in under two minutes.

  • NewA 1 min 41 s film near the top of the homepage: the industry’s biggest problems in sourced figures (KPMG, Flyvbjerg, PlanGrid & FMI, Arcadis, and Egypt’s currency and steel prices), then the platform itself. It plays only when you press play, and phones get a lighter copy.
  • NewUnder the film, everything it says with every figure’s source — in Arabic on the Arabic page — and search engines see it as a video of the product.
v1.94

Trace every material lot from its maker to where it was built in, see what is in store and when it expires, and send reports that read like a professional report — or ask for any report in your own words.

  • NewMaterial traceability: upload the register and each lot shows its chain — maker and country of origin, importer, supplier, purchase order and delivery, certificate and inspection, store, and the element and location it was used in — with its code, so any lot can be followed to the end. Search by code, batch, supplier or place; a rejected lot that was built in, one used before its inspection was approved, and a lot without a certificate are flagged with what to do.
  • NewRaw-material store: each lot’s quantity, place, origin and maker, shelf life and expiry date (given, or worked out from the manufacture date), storage conditions, condition, and its technical and safety data sheets. Expired and soon-expiring lots, materials at the reorder level, lots kept badly and lots without data sheets are flagged, and the store is listed first-expiring-first-out.
  • NewBoth registers are on Materials & Suppliers, in Quality & Safety and Procurement, in the Upload center with Excel templates and samples, in Ask (where a lot came from, what expires soon) and in a new Materials traceability & store report.
  • NewAsk for any report: describe it in English or Arabic — “concrete quality and material traceability for Tower 2” — and the platform picks the chapters and the project you name, shows what it will hold as you type, and builds it.
  • ImprovedEvery report has a cover with the report type and an overall status, an at-a-glance page (each chapter’s headline figure, the key messages, the actions that matter most), a line under each chapter heading saying what it covers, key messages set apart, and a review and sign-off block.
  • ImprovedEvery chapter ends with the decisions and actions its figures call for — most urgent first, with who acts, by when and what is at stake, and whose decision it is. The PDF, Excel and CSV files carry the same, in the report’s language.
v1.93

Search engines and AI assistants find ConstructINS: Google, Bing, ChatGPT, Claude, Gemini and Perplexity read what it is, who it is for and what it does — and every product link in the footer works.

  • NewEvery demo request records where the visitor came from — Google, Bing, ChatGPT, Perplexity, Claude, LinkedIn, an email campaign or a direct visit — from the first page they landed on, and the einsai leads inbox shows it with each request. First-party only: no cookies, no third-party scripts.
  • NewArabic pages search engines can find: every public page has its own Arabic address — /ar, /ar/explore, /ar/updates and each module at /ar/modules/… — linked to its English version, so an Arabic search finds the Arabic page.
  • NewA public page for every ConstructINS module at /modules — what it does, the stage it belongs to, the releases that changed it, and a demo request — so a search for exactly what a module does lands on it.
  • NewA brief for AI assistants at /llms.txt — what ConstructINS is, who it is for, what it does, the systems it works with and the questions buyers ask — built from the homepage’s own copy, so it never says more than the site does.
  • NewSearch engines get robots.txt, a sitemap of the public pages and structured data on the homepage — the company, the site, the product and its FAQ — and the crawlers behind ChatGPT, Claude, Perplexity, Gemini, Copilot and Apple are welcome by name. They redirected to sign-in before.
  • ImprovedThe homepage title and description say what ConstructINS is in the words buyers search for, in English and Arabic, and a shared link shows a card with the product, its headline and its address.
  • FixedThe VoiceINS and RagINS links in the footer and on What’s new went to addresses that do not exist; they open the products now.
Show 93 earlier releases (back to v1.0)
v1.92

Every report, check and answer counts the same decisions — the contractor’s, the developer’s and the whole project’s — and a question with several parts gets every part answered.

  • ImprovedReports: the summary, the chapters and the Excel and CSV files read one snapshot of the decisions — the developer’s included, so a project with only developer decisions no longer reports 0 at stake. A report read from one side says so; one for the whole project lists each side’s decisions apart, with each side’s money.
  • ImprovedHealth and progress say what is missing: a program, progress, cost or quality register — not “no site records” beside the day’s site activity records, which measure activities, not the whole project.
  • ImprovedReconciliation compares the contractor view, the developer view and the whole project four ways — the registers, a report generated for the check, its CSV export and the dashboards — on the same projects, role and workflow status, and flags any difference instead of agreeing at 0.
  • ImprovedAsk answers a question with several parts item by item — contract value, units sold, collected, remaining, overdue, today’s site activities — each with its source and record date, or exactly why it is not available, in English and Egyptian Arabic; the confidence is the share of the question answered.
  • ImprovedWhere a figure comes from: a record entered on the platform is never called demonstration data — a bundled sample, an entered record, one its author marked as a test, and an uploaded file are told apart — and an answer cites the records it rests on, not unrelated ledgers.
  • ImprovedA rule-based answer says it was answered straight from the records — and no longer asks you to turn on a provider that is already configured.
  • ImprovedAI Control: the approval matrix names approvers for the one selected project only — or says none are named yet; voice agents count what has actually run; and each audit log says what it covers and how it is protected.
  • ImprovedThe “?” help buttons are a full 24-pixel target with a mouse and 44 pixels on a touch screen, with a visible focus ring.
v1.91

Choose the model that reads a document — Claude, ChatGPT or Gemini, any model your keys can use — and read it again with another when it is not right.

  • NewRead with, in Scan & Capture: Automatic reads with the model chosen for reading documents, unclear handwriting re-read on the strongest — or choose any model your live providers offer that reads pictures: every Claude, ChatGPT and Gemini model that Admin → AI → Pull all models lists, new releases included, each with what it is good at (the hardest handwriting, speed, cost).
  • NewNot right? After a read, read the same document again with another model in one step. The page, the text and the paperless copies come from the new read, and the screen says which model read it and that you chose it.
  • ImprovedA model you choose reads the document on its own — its closer look at handwriting included — with no quiet switch to another model; if it cannot read the file, you are told.
v1.90

Scan & Capture reads a page as you see it — every field, table, picture, stamp and signature, what it means and where it sits — and takes a closer look at hard handwriting.

  • NewWhat is on the page, in Scan & Capture: a scanned or photographed document is read as a page — each heading, field (its printed label kept apart from the value), table, picture or drawing, stamp, signature, tick box and handwritten note, in reading order, with what it means and what it is for, where it sits, whether it is handwritten and how sure the reader was. The photo shows every part outlined where it sits, and the text is the page read in order.
  • NewHard handwriting, in Arabic and English: the reader knows the words a site writes — rebar, formwork, مستخلص, شدة خشبية, م٣ — and how Egyptian handwriting writes them, Arabic-Indic digits included. A word it cannot settle is marked with its other likely readings, and one tap takes the right one. Parts it is unsure of get a closer look: re-read from enlarged close-ups on the model chosen for reading handwriting, called only when it is needed.
  • NewReading handwriting, a new task in Admin → AI → Model for each task: the most accurate model that reads pictures, chosen automatically — or by hand.
  • ImprovedScans now run on the model chosen for reading documents — they ran on each provider’s default model — and a phone photo is turned upright and sized for the reader before it is read.
  • ImprovedPaperless copies carry the page: the Excel workbook has a sheet listing every part — where it sits, its kind, label, text, meaning, certainty and unsure words — and a sheet for each table found; the Word file and the PDF show the page part by part, pictures cut from the photo, unsure handwriting marked with its other readings, and which model re-read it.
v1.89

Every AI model your providers offer, read again every morning — and the best one chosen for each kind of work.

  • NewModel for each task, in Admin → AI: answers, reading documents, hard calls, quick reads and speech each run on the model that suits them — chosen automatically from every model your Claude, ChatGPT, Gemini and ElevenLabs keys can use, weighing quality, understanding, speed and price, with the reason shown. Choose another for any task; your choice is kept while the provider offers it.
  • NewModels kept current: Verify key reads every model the key can use straight away, and the lists are read again every morning at 06:00 — a new release is marked New and weighed the day it appears. Pull all models reads every provider again whenever you like.
  • ImprovedEvery AI call goes to the model chosen for its work — Ask and the codes on the answers model; scans, tenders and supplier offers on the documents model; submittal reviews and the agents’ plans on the strongest — and when a provider fails, the next one runs its own best model for the same work.
  • FixedGreen and amber states on the admin screens — a key set, a recommended model, a live provider — show in their colors again.
v1.88

From paper to paperless: a scanned document kept as Excel, Word or PDF before it is checked — and every time on the platform in one local time zone and one format.

  • NewPaperless copy, in Scan & Capture and in Ask’s data input: straight after a document is read — before it is checked against the project or filed — keep it as an Excel workbook, a Word file or a PDF. The file holds what is on screen, corrections included: the document’s details, its lines, the rows for each register, the site work, the full text and the photo of the paper, marked as an extracted copy not yet verified. The same step as ManufactINS’s Document Reader.
  • ImprovedEvery time on the platform is shown in one local time zone — Cairo unless the deployment sets another — for every viewer, and in one format: “29 Oct 2026, 14:05 (Cairo, GMT+3)”, dates as “29 Oct 2026”. The AI settings, your own agents, photos, RFQ awards, reports, release notes and emails all write it the same way.
v1.87

Templates from real tender and quality documents: a tender walked from invitation to award, bids scored on best value, an inspection and test plan, and a BOQ checker that catches a total leaving out a section.

  • NewA tender walked from invitation to award, on Tender Management → Walk-through: the stages of a United Nations Invitation to Bid, five synthetic bids taken through opening, the arithmetic correction and the compliance checks, and the award to the lowest price that passed them all — with its performance security, the 25% quantity change allowed at award, and the line it leaves in your tender history.
  • NewBids scored on best value: give a package’s bids a technical score out of 100 and they are ranked technical 70%, price 30%, with a pass mark of 70 — the way a WHO office in Cairo evaluates a renovation. A bid that looks like a gap is still not taken forward.
  • NewAn inspection and test plan for a standby generator, its automatic transfer switch and the emergency board, in the style of a UNICEF tender: ten checks with their acceptance criteria and records, downloaded from the quality card as a register in English or Arabic. Once it is uploaded, the quality screen shows what passed, what must be repeated and which hold points are still closed.
  • NewThe BOQ checker reads a file’s own totals: every subtotal, the total, the VAT and the total with VAT are checked against the lines. A section with no subtotal of its own, left out of a total that adds the subtotals, is named with its rows and its value. A synthetic Arabic finishing BOQ for 20 apartments on the Import screen shows it — its total leaves out EGP 580,000.
  • ImprovedArabic BOQs are read more as they are written: «بنود الأعمال» as the description, «الفئة» as the rate, spellings with or without hamza and taa marbuta. Rate-only items and unpriced lines are listed instead of entering the contract at zero.
  • FixedA BOQ with subtotal rows was counted twice — the subtotals were read as items. They are now checked, not counted.
v1.86

Read a tender before you price it, send its BOQ out for quotations and compare the offers — in plain English and the Arabic of Egyptian sites.

  • NewThe tender reader, on Tender Management: add the tender documents (PDF, Word, Excel or text) and its BOQ, and see the dates with the days left, the money terms, what to submit, the risks with the tender’s own words behind each, the questions to send the client, the Law 182 and federation checks, and the BOQ split into packages for quotations.
  • NewRequests for quotation, on Procurement and Tender Management: a BOQ package becomes an RFQ, the suppliers you already bought the same thing from are suggested, and you download the sheet and the letter to send. Each offer comes back as the filled sheet or its text, and the offers are compared line by line on one basis, against the Egyptian market range.
  • NewA person awards the RFQ, with a reason, and the rates go into the supplier price list. The cheapest offer is not picked on its own: one priced far below the others or the market waits until its scope is checked, and a single offer is never called the lowest.
  • NewLearn the platform: a short track for each part of the team — site, technical office, project managers, tendering and procurement, leaders and administrators — every lesson done on a real screen.
  • NewData, security and AI: a public page with what the platform does with your data and how its AI is used.
  • NewThe homepage lists the proprietary ConstructINS modules, built in-house by Epsilon for construction — easy to connect and light on resources, with a patent application in progress — and Ask answers what they are.
  • NewAsk starts tasks: recover a delay, review a tender or BOQ, compare supplier offers, or improve a workflow with your own agent.
  • ImprovedWord files (.docx) can be added to Ask and to the tender reader.
  • ImprovedThe demo request asks what you need: a demo, a free trial, training for your team or a workflow built for you.
  • ImprovedPlain, everyday English across the platform — program, labor, color, aging, “for example” — and the Arabic of Egyptian sites: اللوحات، الاستفسار الفني، السماح الزمني، طلب الاستلام، قيمة الأعمال المنفذة.
  • FixedCrew categories on the attendance screen read in Arabic.
v1.85

Claude, ChatGPT and Gemini can answer together and agree — and every model your keys can use is chosen in Admin → AI.

  • NewAnswer together: switched on in Admin → AI, every live provider answers the same question with the same records. When they agree, the answer is given at once; when they differ, one of them reconciles the answers against the records and says where they differed. Ask shows which models answered and whether they agreed. Off until you switch it on — it costs one call per provider.
  • NewEvery model, read from the provider: “Pull all models” in Admin → AI lists every model the key can use — Claude, ChatGPT, Gemini and ElevenLabs — so a new release can be chosen without an update. ElevenLabs voices are listed with the Egyptian and Arabic ones first.
  • NewSubmittal review on Standards & Codes: paste a material submittal and it is checked against the Egyptian codes for that material, the contract’s clauses and the BOQ — each check cited, the verdict a draft for the engineer (code A, B or C), and an approval with a failed check sent back.
  • NewRFIs drafted for you, in English and Arabic: for a drawing revised after pricing that no variation names, and for a drawing awaiting approval that the next work waits on — addressed to the Engineer.
  • NewYour own agents, without code: a name, what to look for in plain words, the projects, how often and who hears. Each run reads the project’s records and sends what it finds as an in-app notice; it never changes a record.
  • NewA daily site diary report: the day’s site activity, the crews, what the work under way or starting this week still waits on, and the open items in the quality register.
  • NewA notice to the commercial manager and the quantity surveyor when a drawing is revised after pricing and no variation names it — closed when a variation covers it.
  • ImprovedThe Workforce and Quality chapters of every report read an uploaded project’s own registers — crews, inspections, NCRs, readiness — not only the demonstration projects’.
  • ImprovedMore Egyptian knowledge behind every answer: contractor registration with the contractors’ federation (Law 104 of 1992 — the grades, the documents, the limit for a non-member), the Public Contracts Law 182 of 2018 article by article, how Egyptian public projects have actually run from the lenders’ completion reports, and the Ministry of Housing building-materials bulletin.
  • FixedThe cost baseline follows the Ministry of Housing bulletin of September 2026: hollow block EGP 19,649 per 1,000 and ready-mix C25 EGP 3,549/m³ excluding VAT. Block and concrete prices that passed on the old figures may now be flagged as low.
  • FixedThe delay-penalty cap is stated as Law 182 Art. 48 has it: 10% of the contract value when the delay is within 10% of the duration, 15% beyond.
v1.84

The playbook’s twelve stages are built end to end — seven wait only for your data, and turn to “Working now” the moment it arrives.

  • NewSix new registers — contract clauses, supplier prices, subcontract bids, plant bookings, the drawing and document register, and subcontractor retention — read from a file, a scanned document or any system posting to the webhook.
  • NewThe contract read clause by clause: each clause becomes the term the review weighs; where a clause and the terms sheet disagree, both are shown; a clause the review does not check is listed for a person to read.
  • NewWhen to buy, from your own supplier prices: each material still to order, its order-by date, its price trend and what waiting costs or saves. Subcontract bids leveled against each other and your own past rates — the suspiciously low one named.
  • NewReady for the next three weeks: each activity checked for materials, crew, plant, drawings and inspections — and every morning at 06:00 Cairo time a reminder to whoever answers for a gap.
  • NewThe change nobody priced: drawings on a BOQ item revised after pricing, with no variation naming them.
  • NewSite progress by phone: each thing said on the VoiceINS line is filed on the project — progress and late deliveries only; anything else waits in the inbox for a person.
  • NewCost actuals from Odoo, by hand or every morning, and from any ERP through the webhook — only the actual changes on a cost line. Close-out: the retention still held from each subcontractor, what releases it, and the handover documents still to approve.
  • ImprovedThe playbook says which stages are built and wait only for your data, and turns each to “Working now — on your data” once it is in. Site progress and delay claims are marked working.
  • ImprovedA link about one project — a reminder or a notice — opens with that project selected.
v1.83

The top bar shows the weather and the time where you are.

  • NewThe weather and your local time — in the top bar on wide screens, and at the foot of the side menu on smaller laptops, tablets and phones: the time in your device’s own zone, named; the weather where your network places you, or exactly where you are once you allow your location.
  • NewOpen it for the details — the place and how it was found (your location, approximate from your network, or the Cairo default), the date, and the site time when yours is not Cairo time, since records are dated in Cairo time.
  • ImprovedYour location is used for the weather only: rounded to about a kilometer, never stored, and asked for only when you tap “Use my exact location”.
v1.82

Choose the document type when you scan — 27 construction document types, “Other” in your own words, or let it detect.

  • NewScan & Capture asks for the document type before reading: site (delivery note, measurement sheet, daily report, timesheet, inspection request, lab test report, NCR, snag list, safety report, permit to work, site instruction, material request), commercial (IPC, supplier or subcontractor invoice, PO, quotation, variation order, notice, contract, bank guarantee) and technical (RFI, material submittal, method statement, drawing, minutes, progress report, BOQ) — or “Not sure — detect it”, or “Other” with what it is.
  • NewIn Ask ConstructINS data input, a photo or uploaded file first asks “What is this document?” — the common types as one-tap choices, more types, other, or detect — then reads it.
  • NewTen more document types are explained and filed: lab or test report and snag list (quality), RFI and minutes (risks), progress report, material request (procurement), permit to work or toolbox talk (HSE), method statement, contract or letter of award, bank guarantee or insurance.
  • ImprovedThe type you choose is the one it is filed as; when the document reads as a different type the result says so, and when it agrees it says that too.
  • FixedOn a demonstration project the modules a scan affects no longer show as updated — nothing is saved there.
v1.81

Scan & Capture and the data-input window ask which project a document is for, from the projects you selected.

  • FixedWith All projects (or several) selected, Scan & Capture no longer checks a document against a default project: it lists the projects in your selection and asks which one to file to. With one project selected it is chosen for you, and it can be changed on the result — the check runs again.
  • FixedThe data-input window does the same: with several projects selected it asks for the project before anything is read or recorded, instead of taking the first one.
  • ImprovedOn a demonstration project the scan result says plainly that its records are checked and shown, not saved.
v1.80

The data-input window in Ask ConstructINS reads a photo or PDF by OCR — what it is, the modules it affects, the checked fields — and files it on one confirm.

  • NewTake a photo — OCR: photograph a delivery note, NCR or site report with the phone’s camera, or the laptop’s camera in the window, straight from the data-input window.
  • NewUpload a file — OCR: a photo or PDF uploaded, dropped or pasted into the window is read by the Scan & Capture reader and answered in the conversation with what the document is, the modules it affects and every value checked against the project; one confirm files it. An Excel or text file still goes with your message.
  • ImprovedScan & Capture and the data-input window share one result view, so a document reads, checks and files the same way in both.
v1.79

Scan & Capture explains what a document is, shows the modules it affects, and files it with one confirm when every value checks out.

  • NewA scanned document now says what it is on a construction project — a delivery note, measurement sheet, IPC, variation, NCR, safety report and more — who issues and signs it, and what is checked before it is accepted, in English and Arabic.
  • NewModules this affects: every screen the document updates — Procurement, Site Operations, Quality & Safety, Commercial… — with what changes there, and the screens it is only relevant to.
  • NewEvery value is checked against its register’s rules and the project’s own records before anything is filed. When all is right the screen says so and one confirm files it — no edits needed; only a wrong or missing value opens for editing, and the check runs again as you type.
  • ImprovedA delivery note or invoice updates its purchase order in place — found by the PO number — keeping everything the document does not mention. The order keeps its own quantity and value, and a part delivery leaves it open.
  • ImprovedThe result shows the existing record each value updates and what changes (status ordered → delivered), what filing moves on the project, and the recognized text and lines.
v1.78

Verify key for Claude now says why Anthropic refused a key, instead of a bare “returned 400”.

  • FixedVerify key sends the same request as every Claude call — including the workspace, when ANTHROPIC_WORKSPACE_ID is set — so a key that works for answers no longer fails the check with HTTP 400.
  • FixedWhen a provider refuses a key it knows (no credit on the account, a workspace required), Verify key and the AI health check show the provider’s own reason and the fix — add credit under Plans & Billing, or set the workspace — and never echo the key.
v1.77

A master plan for every project — the whole program, its phases, who owns each activity, and the plan the AI manages.

  • NewMaster plan now opens on every project in the switcher, the demonstration projects included: the whole program from mobilization to handover against its baseline, with progress, slip, the critical path and the milestones — the same progress, slip and forecast finish the Command Center, Planning & Scheduling and the scope bar show.
  • NewPhases: the activities grouped into the project’s phases — or an uploaded program’s own Phase / WBS column — each rolling up its dates, progress against plan and slip.
  • NewBy owner: who owns which activity — the job role, and the person on the project’s roster who holds it — with what is running now and what is late.
  • NewThe AI-managed plan: the project’s recommendations, approved or rejected right there through the same decision workflow and record; what the AI re-forecast on its own and why, from each month’s progress and today’s site data; and every change that moved the plan.
  • ImprovedThe demonstration tower’s master plan carries its live L9 → L10 lookahead at today’s forecast. A change to a demonstration project’s plan is previewed — what it moves and what it does to the finish — and not saved.
  • FixedA phase with work still to do is no longer forecast to finish before today, nor after the project’s own completion — Planning & Scheduling’s milestones for the demonstration projects affected now say so.
v1.76

Ask ConstructINS fills the screen, and the data-input window takes a typed message, a voice note, a spoken conversation, or a document or photo.

  • ImprovedAsk ConstructINS is laid out like a messenger: the chats beside the conversation, a compact header with an Ask / Data input switch, and the message box always at the foot of the screen — no scrolling to reach it. A new chat opens on suggested questions as cards; under every answer, Read aloud and Copy.
  • NewData input, four ways: type it; record a voice note, turned into text you check before sending; talk it through hands-free — it reads back what it will record and records it when you say “confirm”; or upload a document or photo — a site diary, a delivery note, a whiteboard — which is read and previewed the same way. Drop or paste a file, or take a photo on a phone.
  • ImprovedNothing is recorded until what it shows is confirmed, whichever way it was told. A photo or a scan is read by the AI reader only for an account that may use it, within the month’s AI budget; a long document is read up to its first 8,000 characters, and the window says so.
v1.75

Share a report with named people — the monthly report reaches the client inside the platform.

  • NewShare on a generated report: choose the people on its projects who should read it — the client’s representative for the monthly report — and they find it in their Reports, marked with who shared it, and can open and download it. They read all of it, chapters their own screens do not show included, so it is your choice, person by person.
  • NewOnly people whose access covers every project in the report and who open Reports are offered, by name and role; the person it was shared with cannot delete it or share it on; each share and withdrawal is on the action trail and the person is told. Withdraw it and it is gone from their Reports.
v1.74

Final acceptance: access that is revoked ends at once everywhere, and each person reads only the reports their access covers.

  • ImprovedAn account whose access is revoked or has ended is refused by every part of the platform on its next request — every screen and every data service, not only the main ones.
  • ImprovedGenerated reports stay in the company’s book, but each person sees and opens only the reports whose projects and chapters their own access covers — a cost report is not the site engineer’s or the client’s to read. Send a report to someone outside that access as its PDF or Excel file, or share it with them by name (1.75).
  • FixedA chat renamed in the first moments after its first answer keeps its new name on every device.
v1.73

The whole Egyptian code library, page by page: every one of the 96 documents is held on the platform and read, and answers quote the code itself.

  • NewThe 96 HBRC code documents are held on the platform — 1,389 pages, every page read in Arabic and English — so Standards & Codes and Ask ConstructINS search the codes’ own text, not only the catalog.
  • NewAn answer about the codes quotes the requirement and its values from the page that states it and names the document and page; every result opens that page’s text, and the scan itself at that page.
  • ImprovedContents, symbols and reference pages no longer crowd out the pages that state a requirement, and figures typed in either Arabic or Western digits find the same clause. The text is read from scans, so a figure is to be checked on the page — the answer says so.
v1.72

A master plan that shows the whole program against the original, every change that moved it, live — and where its owners change it.

  • NewMaster plan (Win & plan): the whole program of a project as the platform forecasts it, measured against the original plan — planned and forecast finish, slip, progress, critical path, what has changed from the original — with the AI’s pending proposals beside it, updating live.
  • NewEvery change that moved the plan is in its history: edits made on the master plan, updates from files and from the data-input window, replans accepted and decisions applied — who, when, why, and what each did to the forecast finish.
  • NewPlanning, project management and the leadership change the plan there — an activity’s dates, duration, who it is assigned to or what it waits for, or in plain words (“push T3-SLB-13 by five days”) — previewing what it moves before it is written; the change is recorded with its reason and the planning team is told.
v1.71

Chat windows with their own memory, folders and files; a data-input window that records what happened and re-plans; and answers from the codes.

  • NewAsk ConstructINS keeps every conversation in its own window, like ChatGPT: a sidebar of chats named by their topic, renamed, moved into folders and deleted; each window remembers its own turns, so a follow-up is answered in its conversation. Chats are saved to the account and follow you to another browser.
  • NewFolders hold instructions and files that every chat inside them reads, and each chat can have files of its own — PDFs, scans, photos, workbooks and text — read once, with only the passages that bear on a question sent with it.
  • NewA data-input window takes what happened in plain words — “the project manager is off sick today”, “slab L12 is blocked”, “the rebar delivery slips four days” — matches it to the project’s roster, program and orders, shows what it will record and what it moves, and on confirmation records it, re-plans the program, hands the absent person’s work to a named cover and tells the right people.
  • NewStandards & Codes answers questions: the question is read by Claude, ChatGPT or Gemini, the code library is searched by meaning in English and Arabic, and the answer names the code, part, chapter and edition that govern, citing the documents — each opening the scan itself.
v1.70

Claude and ChatGPT read any document into the right registers, and every role agent can plan its day with AI.

  • NewA scan or photo of any document — a daily report, an NCR log, an IPA, a delivery note, a variation — is read by Claude, ChatGPT or Gemini and filed in every register it holds records of, in that register’s own columns; one daily report can feed site progress, quality and safety at once. Each item is checked like a spreadsheet and previewed before anything is written.
  • NewEvery role agent can think its day through with AI: the model reads all of the agent’s checks and puts the next steps in order with the reason for each, pointing only at what the checks found and at actions the account may take — a person still takes every action.
  • ImprovedChatGPT now reads PDFs as well as images; current Claude models are given room to think and the effort each task needs, with the server-side fallback when a request is declined; a document read and its commit are the same read, so what is approved in the preview is exactly what is saved.
v1.69

An AI agent for every job on the project — it checks every record in seconds, drafts the paperwork, and prepares the decisions a person takes.

  • NewEvery one of the 47 job roles has an AI agent (AI Agents): it runs the job’s checks over every record of a project in seconds — the program, the site reports, the quality, safety, cost, payment and procurement registers — and lists what needs action, most urgent first.
  • NewEach agent drafts the job’s paperwork, ready to preview and download: the daily site report, the two-week look-ahead, the NCR chase list, the HSE summary, the cost report, the payment chase letter, draft notices and supplier expedite requests. Letters and notices are drafts a person sends.
  • NewThe QA/QC and HSE agents chase the owners of overdue items inside the platform when you say so — one message each, listing their items — and today’s reminders can be withdrawn.
  • NewAgents prepare the decisions their job may approve — who will carry each one out and exactly what confirming it will change — and the decision stays with a named person. A person with a job role runs only their own job’s agent, and an agent reads only what its person may read.
v1.68

Decisions reach the person responsible, a new project is shown around after its upload, and a large demonstration project loads in one click.

  • NewApproving a recommendation on a project added here names the person responsible from the project’s own roster and notifies them. They confirm it was applied, and the project’s registers are updated and every screen re-reads — NCRs closed, a certificate marked paid, a delivery date committed, an extension determined, an activity started.
  • NewNotifications in the bell are addressed to people by name and job role, only on the projects they can see; whoever approved a decision is told when it has been applied.
  • New“Show me around after upload” on every upload form walks through each screen the new data fills — the menu item, then every tab — saying what it reads from the upload.
  • NewA large synthetic demonstration project — six towers, a retail podium and infrastructure, about 18,000 records — loads in one click from the Upload center, or downloads as the files a client would upload.
  • FixedAccounts limited to their own projects no longer find sample-project records in search; the profile shows the account’s real role; Settings shows workspace-wide user and audit counts to staff only.
  • Improved“Program” is used throughout, and the homepage wording has been revised.
v1.67

Every role sees only its own projects, and every figure keeps its exact value, its source and its unknowns.

  • FixedAccounts limited to their own projects open every screen in their menu — Cost, Commercial, Risks, Handover and Workforce no longer fail — and never receive another project’s figures, names or files.
  • FixedAmounts keep their exact value everywhere: a EGP 250,000 bill stays EGP 250K in headers, Portfolio, reports, exports and Ask; totals are rounded once, where they are shown.
  • FixedReconciliation compares amounts as well as counts — including the developer’s sales, collections and budget — and fails on a difference instead of calling it reconciled.
  • FixedSite records are checked before they are saved — no negative or impossible quantities — and lowering a recorded quantity is a correction with a reason; every change, refusal and correction is in an exportable action trail.
  • FixedAsk, reports and dashboards read the same saved site records, with who recorded them and when; a project with no measured progress or health shows it as not available, never 0% or 100.
  • ImprovedOpen screens update after a change is saved anywhere, and say when their figures were last confirmed.
  • ImprovedBuyer aging without an overdue date is shown as an estimate, never older than the contract; the briefing quotes the current decision record.
v1.66

One set of figures on every screen, report, export and answer — each saying which projects it covers and where it comes from.

  • FixedCash and certification: the Commercial roll-up, Cost, reports, exports and Ask now read one ledger for the projects in scope, so outstanding, overdue and their aging agree everywhere; the reconciliation checks each of those against it, and a mismatch stops the pre-demo check.
  • FixedRisk exposure: one probability-weighted figure on Portfolio, in reports and in Ask, with the gross figure beside it; a late-payment risk is raised only when a certificate is past its payment term, and is shown as prospective otherwise.
  • ImprovedWhole-portfolio reports and exports include projects added on this platform and list every project they cover by code; a project without a register is listed as excluded with the reason.
  • ImprovedProgress, health and status say whether they are measured, imported, calculated, assumed or not yet available; a project with only a BOQ shows its progress as assumed and its health as not yet scored.
  • ImprovedRecommendations show their arithmetic — direct savings, cash timing and losses avoided kept apart — with the options weighed and the risk that remains.
  • FixedQuestions about several selected projects are answered in full, in Arabic or English, with the combined figure; recovery and handover dates name their project and milestone.
  • FixedExports download under the site’s security policy and confirm the file; small amounts show in pounds, never as “EGP 0M”; switching projects never shows one project’s figures under another’s name.
  • ImprovedAsk and document analysis name the AI that actually answered; the pre-demo check reports integrations, mail and voice as they are, marking what still needs a person to verify.
v1.65

Ask answers from each project’s own records, on the AI you choose — and every screen shows only the projects you select.

  • NewEach project has its own knowledge base: every upload, weekly update and linked email or WhatsApp message is added as it arrives. Ask, voice chat and the live call answer from the selected projects only — one or several.
  • NewPredictions and next steps from the project’s data: forecast completion, cost at completion, margin, peak funding and delay damages, with the platform’s recommendations ranked by impact.
  • NewChoose the AI: Claude, ChatGPT or Gemini, in the order you set in Admin, with ElevenLabs for voice — one sector prompt for all of them: Egyptian contracting, FIDIC 2017, Law 182, the Egyptian Code, VAT and withholding.
  • NewWork done but not yet certified is flagged with the amount to put in the next IPA.
  • ImprovedOne margin everywhere — revenue against cost at completion — and the signed contract value wins over the BOQ total.
  • ImprovedSelecting several projects shows each added project’s own data on every screen; sample figures never mix in. Exports of an added project download its workbook, and Decisions shows its own recommendations.
  • ImprovedArabic reviewed across the platform with the sector’s Egyptian terms and correct number agreement.
  • FixedDelay notices: one per event, counted only when the notice names the work or its cause — no notice shown as both done and due.
  • FixedLate deliveries that have not arrived say so; no negative output targets; failed tests that were repaired no longer count as open; VAT shows as 14%.
  • FixedPages no longer break on a browser set to Arabic: numbers keep one format.
  • FixedSample project: the slab held for rebar with nothing above it started, certification tracking progress, and site notes that match each trade.
v1.64

Connections & inbox: the platform wired to your email, WhatsApp, task tools, drives and business systems — replies suggested from the project data, sent only when you approve.

  • NewEmail: any company mailbox over IMAP/SMTP (cPanel, Zimbra, Exchange, or Gmail and Outlook.com with an app password), Gmail through Google sign-in, and Outlook / Microsoft 365 through Microsoft sign-in. Messages are pulled into one inbox, linked to their project by its code, name and references (IPA, PO, NCR, variation and activity numbers), and each gets a suggested reply.
  • NewWhatsApp Business (Cloud API) receives the messages sent to the company number through a signed webhook and sends approved replies; Telegram, Slack and Microsoft Teams channels too.
  • NewSuggested replies answer from the project’s own figures — payments, deliveries, progress and forecast, open NCRs, pending changes — in the sender’s language. With the AI provider on, the reply is written by the model from a fact sheet and told never to invent figures; without it, the platform recognizes what is being asked and answers with the same facts. Every reply waits for approval — queue it, or approve and send after a confirmation — and every approval is logged.
  • NewTask tools: ClickUp, Jira, Asana, Trello, monday.com and Odoo — create tasks from drafts and actions, and see their status here.
  • NewDrives and Microsoft 365: Google Drive (Google Sheets and Docs come in as Excel and Word), a Drive folder shared by link, OneDrive and SharePoint libraries, and Dropbox — browse folders, import a program, register or report straight into a project’s update, and save the project workbook back.
  • NewERP and CRM: Odoo purchase orders into a project’s procurement and HubSpot deals; SAP, Oracle, Dynamics 365, ERPNext, Salesforce, Zoho, Primavera P6, MS Project, Aconex, Procore and Autodesk Construction Cloud through the platform’s signed webhooks — records posted to a secret URL are read exactly like an uploaded file, and the platform’s events (sent messages, created tasks) are pushed out signed with HMAC-SHA256 for Power Automate, Zapier, Make or n8n.
  • ImprovedMessages & letters can now go out through a connected service (queued for approval) or become a task in a connected tool.
  • ImprovedPasswords, tokens and sign-ins are stored encrypted per company and never shown again; nothing is sent from the platform without a person approving it.
v1.63

Try everything on a real-looking Egyptian project, send the week’s letters and messages from the data, and take the project back out in Excel.

  • NewSample project in the Upload center: Yasmeen Heights Residences, Tower 2 — two basements, ground and ten floors in New Cairo, a 30-month job about sixteen months in — as the files a contractor and its developer keep: priced BOQ, contract terms, Primavera-style program, cost plan, purchase orders, a 247-person site roster with subcontractor gangs, daily site reports with quantities, inspections, cube tests and NCRs, variations, notices and claims, interim payments net of retention, advance recovery, withholding and VAT, handover documents, risk register, HSE register, the developer’s budget and unit sales, a company tender history and a weekly update workbook. Load it in one click to see every screen fill in, or download the files (one by one or as a .zip) and upload them yourself. Dates move with the day you download, so it always reads as a live job.
  • NewEvery upload card has a “Sample” file beside its template — a filled-in example ready to upload — and the Update a project tab offers the sample weekly update workbook.
  • NewMessages & letters (Contracts, Commercial, Procurement, Quality & Safety, Handover, Decisions and the Upload center): the week’s correspondence drafted from the project data — notices of delay inside their time bar, overdue certificates with their figures and the contract remedies, requests to decide pending variations and claims, supplier chases naming the activity a late delivery holds, NCR actions to the person responsible, blocking handover items, and the weekly progress report. Each in English and Arabic as a letter, an email and a WhatsApp message; edit, then copy, open in email, open a WhatsApp chat (with the roster’s number when it has one) or download the letter as a Word document. Nothing is sent from the platform.
  • NewProject workbook (.xlsx): the project as the platform has worked it out — summary and observations, the program re-forecast with float and critical path, each person’s lookahead, the monthly cash flow with VAT and withholding, contract deadlines, contract review, replan options, late materials, quality hold points and risks — in English or Arabic.
v1.62

The official Egyptian code documents are now in the platform — searchable, readable in place, and cited by Ask ConstructINS.

  • NewEgyptian code library (Standards & Codes): 96 HBRC code documents from the company’s Drive, grouped under 15 codes — concrete structures (ECP 203, including the full 2018 update), loads (ECP 201), soil mechanics and foundations (ECP 202, all ten parts and the 2006 guidance manual), steel structures, building (masonry) works, electrical installations (ECP 302) and its guidance manual, elevators (ECP 303), sanitary installations (ECP 301), water and sewerage pipelines, treated wastewater reuse (ECP 501), urban and rural roads, road coordination, accessibility (ECP 601) and hospital design standards.
  • NewOpen any document in place — the scanned pages beside what it covers, its topics, its decree and edition — or in Google Drive. Each part shows its current edition first; earlier editions it replaced are kept and marked. Files that hold only a part’s cover, decree and contents say so.
  • NewOne search box covers the codes and the documents, in English or Arabic, across code names, parts, chapters, decrees and topics. A link such as /standards?doc=K5 opens a document directly.
  • ImprovedAsk ConstructINS and the AI consultant now know the library: a question about a code topic cites the document it comes from, with a link that opens it, whatever projects are selected.
v1.61

Plans run on Egypt’s working calendar, cash flow carries Egyptian VAT and withholding, and public contracts are reviewed against Law 182.

  • NewEgyptian working calendar for every program: a six-day week with Friday off, Egypt’s public holidays (the national days, Coptic Christmas, Sham El-Nessim, both Eids, the Islamic new year and the Prophet’s birthday) and reduced output in Ramadan. Durations are working days, so forecasts skip rest days and holidays and stretch work in Ramadan. Planning lists the holidays and Ramadan still ahead, the working days each costs and the critical activities it falls on; the rest days, holidays and Ramadan output can be changed per project under Dynamic replan.
  • NewCash flow with Egyptian taxes: 14% VAT on each certificate, paid to the Tax Authority on the next month’s return (less input VAT) before a slow client pays it, and 1% withholding deducted from every receipt. The funding peak includes them, and the screen shows the VAT carried for the client at its peak and the total withheld.
  • NewContract review for Egypt: public works under Law 182 are checked against the 5% final guarantee due within 10 working days of award and the statutory delay-penalty scale up to 15%; every contract is checked for a change-in-law and taxes clause, flagged high where the contract was priced before construction moved from the 5% schedule tax to 14% VAT in 2025.
  • ImprovedTender proposals now state that prices exclude 14% VAT, that the program allows for Egyptian holidays and Ramadan, that changes in law and taxes adjust the price, and — for public bodies — the Law 182 guarantee and price-adjustment indices to confirm.
  • FixedUploads are hardened for production: an oversized or malformed workbook can no longer exhaust memory or stall the server, program and register files are read in bounded time, a damaged project file is set aside instead of being overwritten, and one unreadable file no longer fails a whole update.
  • FixedA login limited to some projects now sees tender intelligence learned only from the company history and its own projects, and only logins with access to every project can change the company tender history. Requests from other sites to change data are refused, and project updates up to 40 MB go through.
v1.60

Update projects with any files, price tenders from the company’s history, review contracts, and plan the money, the people, the deadlines and the quality.

  • NewUpdate a project with any files (Upload center → Update a project). Drop a monthly report workbook, an updated program, progress sheets, the IPA register or the NCR log. Each sheet is recognized, then checked on its own and against what the project holds — progress going backwards, reports on activities the program does not have, more paid than certified, dates out of order, actual dates in the future — and compared row by row and field by field. Before anything is applied the screen shows what the update moves: progress, forecast completion, slip, cost at completion, delay-damages exposure, health, open NCRs and cash. Every update is logged with its effect.
  • NewTender intelligence (Tender Management). Upload the company’s tender history — what was bid, the margin, the winning price, won or lost, and for won jobs the final margin, overrun, delay, claims and lessons. With the platform’s own projects as outcomes, each tender gets the win rate of similar tenders, how far above the winner the lost bids were, a win chance fitted at every margin, the margin with the highest expected profit, a contingency from how similar jobs overran, a program buffer, history with the same client, the lessons that apply, and the cases a proposal must cover — downloadable as a proposal summary.
  • NewContract review (Contracts). Every key clause — payment period, financing charges, retention, advance, delay damages and their cap, price adjustment, exchange rate, force majeure, liability cap, performance bond, claim notice period, suspension, termination for convenience, defects liability, disputes — checked against the protections a contractor needs, with the position to negotiate, the money at stake and the margin left after contract risk.
  • NewFinancial and operational plan (Planning, Commercial, Cost). A monthly cash flow from the program, budget and payment terms — cost out, certified, cash in after the payment period, retention and advance recovery — with the peak funding need and when the job turns cash-positive, and the actual receipts beside it. A 12-week resource load: who is busy, who is free, who is double-booked and which work has nobody on it — and a named suggestion for each unassigned activity, matched on skill, availability and how each person has delivered so far.
  • NewContract obligations and deadlines (Contracts, Commercial, Handover). Notices of delay with their time-bar date from every blocked progress report, decisions to chase on pending variations and claims, payments due on each certificate, the next interim application, milestones, retention releases and items required for handover — overdue first.
  • NewQuality and rework (Quality & Safety, Site Operations, Handover). Rework rate and cost by trade, first-time pass rate by month, a hold-point inspection before every activity starting in the next three weeks in a high-rework trade or on top of open NCRs, and when open snags will be cleared against the handover date.
  • ImprovedThe data-sources strip on a project brought in from its files now lists those files, with the date each was uploaded, instead of “sample data”.
v1.59

Bring any project in from its own files — new from tender, or already on site — and every screen fills in.

  • NewUpload center (Portfolio → Upload center). Start a new project from its priced BOQ or just its key facts, or bring in a project already on site. Then upload its files as you have them, on tabs that follow the project: tender & contract terms, variations, claims & EOT, program, procurement, budget & cost to date, interim payments, workforce, site progress, quality inspections, NCRs & snags, handover checklist and risk register.
  • NewReads the files people already keep: Excel workbooks (.xlsx, every sheet tried), CSV, Primavera P6 (.xer) and MS Project (.xml) programs, and PDFs or photos of a register through the AI reader. Headers can be English or Arabic, in any order, under a title block. Dates are read day-first, amounts with EGP, thousands separators or “1.2M”, and percentages as 45, 45% or 0.45. Every file is previewed first — matched columns, skipped and repaired rows, the first rows as stored — and nothing is saved until you confirm. Replace or add to what is there; each kind keeps its upload history. Templates for every kind are one click away.
  • NewProject tracking from the uploads. The program is scheduled with its logic (finish-to-start, start-to-start, finish-to-finish, lags), re-forecast from the status date with actuals fixed, and compared with the contract completion: % complete against planned (SPI), slip, critical path, float, activities behind or late to start, and how much runs in parallel. Budget and cost to date give earned value, CPI and cost at completion; contract terms give delay-damages exposure (capped), retention and the defects-liability end; interim payments give the cash position and what is overdue.
  • NewA work plan for every person. The roster and the program’s assignments (names, employee IDs or crews) become each person’s next three weeks — today’s tasks, % complete against planned, and whether they are on track, behind or free. An uploaded roster also staffs the Crew board, so attendance works for those people from day one.
  • NewDynamic replanning: late starts, work going too slowly at the rate actually achieved, overlap options re-forecast before you accept them, people on too many activities with a named colleague free to take one, materials arriving after the work needs them, old NCRs, and what each day of slip costs. Accept a recommendation and the whole program re-forecasts at once; undo it as easily. Move the status date to re-measure.
  • ImprovedEvery module now shows an uploaded project’s own data instead of “No records yet”: Planning shows its program, replan and plans; Cost, Commercial and Contracts its budget, cash, contract and changes; Workforce and Crew its roster and plans; Site Operations its progress; Quality, Handover, Procurement and Risks their registers. Where a file has not been uploaded the screen names it and links to it. Portfolio figures for the project (progress, forecast, slip, health, headcount, open risks) come from the uploads, and the sample header figures are hidden for it.
  • FixedSite photos, photo requests and field updates are accepted for projects added on the platform, not only for the demonstration projects.
v1.58

Photograph a document with your phone, tablet or laptop camera and read it straight away.

  • NewScan & Capture now takes document photos with the camera on a phone, tablet or laptop. Before the browser asks, the screen explains why the camera is needed. If access is already allowed the camera opens straight away. If it is blocked, the screen shows how to unblock it on that device: iPhone/iPad, Android or laptop.
  • ImprovedThe camera uses the rear lens on phones and tablets and the built-in camera on laptops. It has a page guide, a switch-camera button and the torch where the device has one. Each shot gets a sharpness and lighting check, so a blurred or dark page is retaken before OCR rather than after. Photos keep up to 2200 px, so small print stays readable. With no https or no camera, the device's camera app or a photo upload is offered instead.
v1.57

Work that overlaps is planned as overlapping — not forced into a single line.

  • ImprovedThe live lookahead now plans overlaps the way a site runs them: an activity can start a set number of days after another starts (start-to-start) or finish with it (finish-to-finish), not only after it finishes. Rebar fixing on L10 follows the formwork a day behind it instead of waiting for the last panel, which brings the slab pour forward.
  • ImprovedEach lookahead row says how it is tied in — "after Pour L9", "starts with Formwork L10 +1d", "finishes with Formwork L10" — and how many activities run alongside it; the header shows the most activities in parallel at once.
  • FixedThe delivery lifecycle no longer marks Plan & procure "complete" while packages are still being bought, or Certify & get paid "ahead" while interim applications are already moving. Stages next to the current one that still have live gates now show as running in parallel.
v1.56

Every card answers when you press it, and the Command Center is yours to arrange.

  • NewCustomize the Command Center: choose the key figures your role panel leads with — from cash at stake, margin and portfolio value to open risks, overdue receivables, weighted health or worst slip — put them in order with the first leading, and show, hide or reorder every section and both columns. Saved in your browser per role focus; Reset restores the default.
  • ImprovedEvery KPI tile is interactive: hovering shows where it goes, pressing opens that screen. A tile with nowhere else to go — or one already on its own screen — opens a short explanation: what it means, how it is worked out and what makes it up, with links into each part. The decision queue's cash on the table, act-now count and mean confidence work this way.
  • ImprovedCommand Center panels open the screen they summarize — the briefing and tasks the decision queue, the health matrix and status the portfolio, risks the risk register, the health strip the portfolio and site operations — and each key metric opens its own screen.
  • ImprovedCalmer tiles: a copper edge and soft wash on hover, with the destination shown along the bottom so it never covers the label or the figure.
v1.55

Every release, on the public News & updates page.

  • NewNews & updates now lists every ConstructINS release — what was added, improved and fixed — under the latest news, with a link to news from every InsAI platform.
  • FixedDates in the news bar and on News & updates no longer show a day early for visitors in the Americas.
v1.54

Charts that fit their data, and KPI rows that lead with what matters.

  • ImprovedEvery KPI row now leads with its most important figure at display size — forecast at completion on Cost, claim exposure on Contracts, pending variations, superseded drawings still in use — with the supporting measures beside it.
  • ImprovedEvery KPI says what it means: tiles without their own caption now show a plain definition, in English and Arabic.
  • ImprovedCharts are drawn at the card's real width on round-number gridlines, so text no longer stretches; the main line carries its latest value, and hovering any bar or point shows its exact figure.
  • ImprovedBudget performance is now a bullet chart against the approved-budget line, so the forecast overrun is visible instead of capped at a full bar.
  • FixedColors mean one thing: categories (cost elements, exposure types, where a tower's money goes) use a color-blind-safe identity palette, and red/amber/green appear only where they mean good, attention or critical. Exposure by category is ranked largest first.
  • FixedSteel price and demand pressure are plotted as indices from February = 100 instead of two different measures on one axis; crew productivity and fatigue risk are two solid series on one % scale; the delay-risk heat map uses one hue from light to dark with a scale.
v1.53

A calmer, enterprise-grade interface — and decisions that move through a governed approval workflow.

  • NewEvery AI recommendation now moves through a governed workflow — Detect, Analyze, Recommend, Review, Approve, Execute, Verify — with each step attributed and kept in an audit trail. Only the approver the decision names can approve it, rejecting needs a reason, and verifying needs the actual result. Steps are saved, so an approval no longer resets when the page reloads.
  • NewRecord details: selecting a row in any module table opens a side panel with every field, a shareable link to that exact record, and previous/next buttons (or the [ and ] keys).
  • NewReporting period in the top bar — last 3, 6 or 12 months, or full history — sets how much history every monthly trend chart shows. Point-in-time figures keep their own date.
  • NewKeyboard shortcuts on desktop: G then a letter jumps to a screen (G D for Decisions, G C for Command Center), / opens search, and ? lists them all.
  • ImprovedThe Command Center leads with one figure for the role you are viewing as — cash at stake for a CEO — with three supporting figures and named next actions, instead of the same portfolio numbers repeated three times.
  • ImprovedDecision cards separate what was observed, what is forecast if nothing is done, what is recommended and what acting is estimated to be worth, and show the owner, approver, deadline and data date. The evidence now reads in Arabic too.
  • ImprovedQuieter, more legible screens: one compact project-scope line on module pages, copper reserved for actions and the current selection, darker status colors for readable text, theme / language / interface style gathered into one Display menu, and links named for where they go.
  • ImprovedPhones and tablets: KPI names no longer cut off, figures stay on one line, module tables become two-line records, and the Ask button shrinks to an icon so it no longer covers content.
v1.52

Protected forms and sign-in, and a refined visual system.

  • ImprovedThe demo-request form and sign-in now carry the same protections as ManufactINS: automated and repeated submissions are screened out in the background, and repeated wrong passwords slow down and lock the account for a while — with nothing extra for a genuine visitor to do.
  • ImprovedDemo requests are delivered to the Epsilon team signed, so a request can be verified as genuinely coming from this platform.
  • ImprovedA refined visual system: one consistent type scale across every screen, sentence-case labels instead of shouting capitals, a calmer navigation rail, and KPI tiles that keep figures in ink with a small status dot.
v1.51

Report from site by talking to Ask — and capture photo evidence that is checked as it is taken.

  • NewField input through Ask ConstructINS: type, speak or say it on a live call — "Mahmoud started at 7, slab pour L10 finished, ductwork blocked waiting for GI duct" — and it is read into attendance and progress updates for you to confirm, in English or Arabic. Confirmed updates are written to the site record and the live lookahead replans straight away, raising or clearing decisions.
  • NewPhoto Evidence & QA: managers and supervisors can request photos, and the system issues random spot checks during the day. Workers take phase photos before and after each stage, the camera checks light and focus as they shoot, and each photo is stamped with time and place.
  • NewComputer-vision checks (when an AI provider is connected): progress is estimated from the photo and compared with what was reported, and finished work is inspected against a quality checklist, with a large gap raised as a decision.
v1.50

Every account sees the screens its job needs, and the menu shows every module.

  • NewJob-role accounts: executive, project director, project manager, site engineer, foreman, site worker, quantity surveyor, planner, procurement, HSE and client — each with its own home dashboard, its own set of screens and its own permissions.
  • NewMy Day for site workers: a phone screen to confirm the start of the shift, each task and the end of the day — and nothing else.
  • ImprovedThe navigation rail gives its full height to the modules, with every group open, a "Find a module" search and pinned favorites. Settings and Admin are merged into one place, reached from a single workspace menu at the foot of the rail.
  • FixedScreens no longer run off the edge on phones and small tablets; wide tables and charts now fit their column.
v1.49

Scan documents into the site record, and track crew attendance with escalation.

  • NewScan & Capture: upload any document or photograph one with the camera — a delivery note, an invoice, a measurement sheet — and its reference, date, supplier, line items and totals are read out (OCR) for review, then added to the day's site capture in one step. Reading needs a connected AI provider.
  • NewCrew & Attendance: every worker's shift for the day, confirmed from the field (phone or CCTV verification) — start, each task and finish. A missed start escalates to the supervisor after 15 minutes and to the project manager after 45, and the board shows labor availability and trade gaps for planning, cost and procurement.
  • FixedThe floating Ask window now answers only from the projects selected in the switcher, the same as the full Ask screen.
v1.48

Paper-free site reporting on tablets, and a data-integrity pass across the platform.

  • NewField capture on tablet: site engineers record the day's activities — executed quantities, progress, crew, plant, blockers and notes — and the site dashboard updates live from the same record: daily productivity, progress, completion, output by unit and zone, and bottlenecks.
  • FixedFigures now agree across screens: steel priced in EGP per ton, the portfolio margin derived from the projects themselves (about 11.6%), each project's cost decision quoting its own numbers, retention held at EGP 20.6M everywhere, and approved versus unsigned variations matching the commercial register.
  • ImprovedHonest labeling: autopilot performance figures are marked illustrative on sample data, and lifecycle stages are derived from each project's real status and progress.
v1.47

Ask answers from the project records.

  • ImprovedAsk ConstructINS now answers from the project records themselves — and says plainly when a record is not available, rather than answering about the project in general.
v1.46

See the real platform before signing in.

  • NewThe public Explore page shows the actual platform screens on sample data — captioned and enlargeable — so a visitor can preview the product and its functions without an account.
  • ImprovedThe country selector is on the Explore page too, so contact details match the visitor's nearest office.
v1.45

A guided, Epsilon-managed demo journey and a more credible homepage.

  • NewA managed demo journey on the Explore page: request a guided demo, an Epsilon specialist walks you through on sample data, then issues time-limited individual credentials. The form requests a demo; it does not create an account.
  • ImprovedA simpler demo form — phone and WhatsApp optional, country and website optional — with a clear note on how your details are used.
  • NewThe homepage now shows why Epsilon (founded 2015, ISO 27001/9001/21001 and ITIDA accredited, offices in Egypt, Saudi Arabia and the USA) and signed client recognition letters, clearly labeled as Epsilon company experience. The hero figure is labeled as an illustrative sample.
  • NewThe owner can create additional full-access administrator accounts from the console; passwords are stored hashed and shown only once.
  • ImprovedSharper, high-resolution product screenshots, with sample company names blurred; images open larger in a viewer.
v1.44

Security hardening for public access.

  • FixedThe shared sample account can no longer sign in to the production platform; evaluation access is issued individually by Epsilon.
  • FixedWhen access is revoked or expires, data export and Ask ConstructINS stop working immediately, not when the session would have ended.
v1.43

Product showcase, a live news feed and smoother navigation.

  • NewThe homepage showcases the most important modules — real screens paired with what each one does for a client — and they enlarge on click.
  • NewA news bar and a News & updates page bring the latest Epsilon InsAI news for ConstructINS and across the product family.
  • NewA "Platform homepage" link inside the signed-in workspace, and a back-to-home link on the sign-in screen.
  • ImprovedConsistent product naming across the family, a fixed-width language toggle, and an ecosystem band linking every Epsilon InsAI product.
v1.42

Reach your nearest office from anywhere in the platform.

  • NewA region control at the top of every page: choose your country and every email, phone and WhatsApp contact switches to that office, with a featured "your office" card.
  • ImprovedThe homepage lists every Epsilon InsAI product and links to the module showcase.
  • FixedSigning out and signing in now always return you to the right page.
v1.41

Consistent demonstration dates, and the Explore page and guided tour in Arabic.

  • FixedDemonstration dates now follow one snapshot (September 2026): payment applications, project starts, the focus project's milestones and the steel buy window no longer sit in the future. Forecast finish dates stay ahead, because a forecast has not happened yet.
  • ImprovedThe public Explore page and the guided tour read in Arabic.
v1.40

Answers end cleanly (even after "e.g."), a reconciled cash ledger, and honest readiness checks.

  • FixedAI answers no longer end mid-sentence on an abbreviation or open bracket ("…a specific job (e.g."): the truncation trimmer now ignores abbreviation dots (for example, that is, etc.) and unbalanced parentheses when finding the last complete sentence, and the token budget is larger. Tests cover it.
  • FixedThe cash ledger now reconciles on screen: interim payment applications show certified, paid-to-date and remaining balance, and receivables aging is measured on the remaining balance — so the buckets total the outstanding EGP 74.5M and the overdue EGP 31.2M, and a certified figure larger than outstanding is explained (most of it is paid).
  • ImprovedThe reconciliation report and the pre-demo “metrics reconcile” check now cover certification, receipts, outstanding, overdue, the aging buckets and each IPA’s certified/paid/remaining balances — not just counts and headline totals.
  • ImprovedThe Arabic readiness item no longer asserts the deployed site is translated (it does not read the live site); it reports what the source and the render scan verify and tells you to confirm on the running build, since stale caches can show English.
v1.39

No more cut-off answers, Arabic on every client screen, and a stricter style CSP.

  • FixedAI answers no longer end on a colon, comma or mid-word: a reply the model cuts off at the token limit is now trimmed to its last complete sentence (or, failing that, closed with a period), the budget is larger, and tests assert answers always end on complete punctuation.
  • ImprovedThe remaining client-facing screens are now in Egyptian Arabic too — the playbook, the Egypt cost baseline (benchmarks), standards & codes, forex, news, data & insights and reports — including chart labels and the long narratives. Official code identifiers (ECP, ASTM, Eurocode), law numbers and company/project names stay in Latin by design.
  • ImprovedThe readiness note is now precise about Arabic: it lists the translated client-facing screens and states that the internal staff/admin screens (AI control, readiness, reconciliation) are intentionally left in English.
  • ImprovedContent-Security-Policy: the base style-src no longer carries ‘unsafe-inline’ — it is confined to style-src-attr (the one place React’s dynamic style attributes require it), with style elements locked to same-origin or the request nonce.
v1.38

Arabic across the module screens, no more cut-off answers, and tighter style CSP.

  • FixedAI answers can no longer end mid-word: a longer reply budget, and any reply the model cuts off at the limit is trimmed back to its last complete sentence.
  • ImprovedThe signed-in module screens (portfolio, cost, procurement, workforce, risks, handover, decisions, agents, lifecycle, command-center) now read in Egyptian Arabic — section headings, chart and KPI labels, status words, stage and agent names, and the AI decision, risk and handover narratives. Company and specific project names stay in Latin by design.
  • ImprovedThe readiness note no longer overstates Arabic coverage; it now describes exactly what is translated and what stays Latin.
  • ImprovedContent-Security-Policy: stylesheet elements are now restricted to same-origin or the request nonce (style-src-elem), leaving inline styling to the attribute fallback — a step tighter for an enterprise security review.
v1.37

Conversation history per project, live voice transcripts, and voice scoped to the selection.

  • NewA History tab: every Ask ConstructINS chat and voice call is saved and grouped by the project scope it was held under, each named automatically by its topic. Open one to reload it (and restore its project selection), rename it, or delete it.
  • NewThe live voice call now shows the running transcript — both what you say and the assistant’s replies — as you talk, and saves it to History.
  • ImprovedA voice call is now confined to the current project selection, exactly like the typed Ask: the agent is told which project(s) are selected and answers only about those (or the whole portfolio when nothing is narrowed).
  • ImprovedSelecting a project (one, several, or all) drives the typed Ask, the knowledge base and the voice call together, and each new selection starts its own saved conversation.
v1.36

Voice privacy specifics surfaced, and restricted-access & voice-vocabulary coverage.

  • ImprovedThe call setup now has a “Data & privacy” panel with the concrete specifics from ElevenLabs’ own published documentation — data-residency selection (EU available), the Data Processing Addendum, the subprocessor list, and the deletion-request contact — shown before the call, so the disclosure is verifiable rather than a promise.
  • ImprovedAdded automated coverage for a restricted client’s access (a company grant confined to its own projects and screens, signed cookie and all) and for the voice agent handling mixed Arabic/English construction questions with figures and project names.
v1.35

Nonce-based CSP, audio-not-retained voice calls, and one voice-start experience.

  • ImprovedThe Content-Security-Policy is now nonce-based with strict-dynamic and no ‘unsafe-inline’ on scripts. A per-request nonce is minted by the server, stamped on every script, and enforced by the browser — verified end-to-end — with the previous static policy kept only as a safe fallback.
  • ImprovedVoice calls are now private by configuration, not just by promise: the ElevenLabs agent is set to not save the call audio and to retain no transcript after the call (record_voice off, zero-day retention). Every voice notice, the AI Control guardrail and the agent’s opening line say the same, now-true thing.
  • ImprovedTalking to ConstructINS starts from one setup dialog — same title, language choice and privacy notice — whether the live call or the built-in voice chat runs underneath, so there are no longer two different-looking entry screens.
  • ImprovedRestricted-client scope (a company grant’s projects and screens) and the voice agent’s construction/QS instructions now have automated tests, so a regression is caught in CI rather than only in a manual walkthrough.
v1.34

Cash answers, benchmark provenance, Arabic grammar, one voice-privacy notice, and clearer voice controls.

  • FixedThe AI can now answer cash-position questions when the live provider is on: certification, receipt aging, DSO and the interim payment applications are in its context, so “what is my cash position and what is overdue?” is answered instead of reported unavailable. Every promoted question now has an automated test proving it is answerable.
  • FixedAI provenance is separated and dated: project facts, sample demonstration data and external market benchmarks are kept distinct, and the model must cite a named source and as-of date for any market benchmark rather than asserting one from general knowledge.
  • FixedArabic number–noun agreement is now correct — “14 مشروعًا”, not “14 مشاريع” — across every count, while possessives like “مشاريعك” are left untouched.
  • FixedOne voice-privacy notice everywhere: the call modal, the in-call panel, the AI Control guardrail and the agent’s opening line now say the same thing — the call is handled (and may be retained) by ElevenLabs, and ConstructINS does not store the audio — shown before and during the call.
  • ImprovedVoice controls are named for what they do: “Stop listening”, “Interrupt reply” and “End call” — so ending your turn can’t be confused with ending the call.
  • ImprovedExpanding an AI answer no longer repeats its summary — you see the concise summary or the full reply, not both.
  • ImprovedThe public sample screens are labeled as a dated illustrative snapshot, so their figures are never mistaken for the live application. HSTS is now emitted once, by the TLS layer.
v1.33

CSP restored, accurate AI readiness, rendered Markdown answers, and a fully Arabic public site.

  • FixedThe Content-Security-Policy header is back on every route. Next 16 renamed middleware to “proxy” and no longer lets it set response headers on rendered pages, which had silently dropped the CSP — it is now delivered as a static header from next.config, with object-src, base-uri, frame-ancestors and form-action all locked down.
  • FixedAI readiness no longer reports the provider unreachable when it is not: the health check now uses the same 30-second budget as a real Ask (a one-token ping can still take ~16s cold), retries once, and reports the measured latency.
  • FixedLive AI answers render their Markdown — headings, bold, bullet and numbered lists, and tables — instead of printing raw “##” and pipe characters. Rendering is to safe elements, never raw HTML.
  • FixedThe public marketing site is now fully Arabic when Arabic is selected: navigation, hero, pilot evidence, product and governance copy, and buttons — with construction terminology in Egyptian Arabic and company/standard names kept as-is.
  • ImprovedExports show a “Preparing export…” state with a disabled button while the file is assembled, then a clear completion or a failure you can retry — so a slow export never looks like a dead button.
v1.32

Consistent project panels, program dates in AI context, and an accurate provider badge.

  • FixedThe Project 360 tabs now belong to the flagship demonstration project only. Selecting any other project shows its own header figures with “no records yet” panels, instead of the flagship’s cost curve and progress under a different project’s name.
  • FixedThe AI now knows each project’s program: baseline start, target completion and schedule slip are in its context, so it can answer “when does this finish?” instead of reporting the date unavailable.
  • FixedAn answer’s language now follows the question’s own script — an English question is answered in English even when the interface is set to Arabic (and vice-versa).
  • FixedThe Ask console badge and scope line now reflect reality: it reads “Live · Claude” only when the provider is actually live, and names the selected project(s) rather than always the whole portfolio.
v1.31

Workspace-scoped AI keys, tighter record handling, and one CSV download path.

  • FixedAI providers that require a workspace now work: set ANTHROPIC_WORKSPACE_ID and it is sent as the anthropic-workspace-id header on every call, clearing the HTTP 400 that forced built-in answers and disabled document analysis.
  • FixedAsking about a specific record (an invoice, certificate or claim) — even phrased as “tell me about invoice …” — now reports it unavailable instead of returning the project’s summary.
  • FixedEvery “Download CSV” (individual reports included) now uses the same proven download path as the working module export.
  • FixedThe Project 360 header no longer shows one project’s figures when no single project is selected — it prompts you to pick one, so the header and the panels can’t disagree.
v1.30

Clearer AI errors, complete multi-field answers, and firmer import validation.

  • FixedWhen the AI provider rejects a call, readiness (and the logs) now show the provider’s exact reason — for example, an unavailable model id — so it can be corrected instead of a bare “failed”.
  • FixedA broad question (“name, value, forecast, progress, health, completion date” or “tell me about this project”) now answers every field for the selected project, in the question’s language.
  • FixedAsking about a record the selection does not hold (a non-existent invoice) says it is unavailable instead of returning the project’s description.
  • FixedEditing a project’s dates on the import screen now clears the preview and the confirmation immediately (even through the date picker), so a stale preview can’t be added.
v1.29

The platform no longer overwrites a dashboard-tuned voice agent.

  • FixedA live call no longer reverts a voice agent you configured in the ElevenLabs dashboard (voice, TTS model, LLM, prompt). Automatic syncs are now non-destructive — they only keep the per-call language override enabled — and the platform pushes its own full config only on an explicit Admin → AI “Sync”.
v1.28

Complete scoped answers, instant human read-aloud, and honest AI readiness.

  • ImprovedRead aloud now streams a natural human voice from ElevenLabs and starts almost immediately (low-latency Flash model, progressive playback); starting one reading stops any other, and pressing the button again stops it instantly.
  • FixedA scoped Ask answer now names every field you asked for (progress AND forecast, etc.) for every selected project, fully in the question’s language.
  • FixedAsking about a record the selection does not hold (a non-existent invoice) says it is unavailable; asking about a project you did not select says it is outside the selection — instead of quietly answering with another project’s figures.
  • FixedAI readiness is now based on a real test call to the provider, so it agrees with how Ask and document analysis actually behave (configured-but-failing no longer reads as “live”).
v1.27

Ask now strictly obeys the selected projects — retrieval, answers and citations.

  • FixedAsk AI is now confined to the projects you select: retrieval, calculations, the built-in fallback and the citations all use only the selected projects (multi-select included). A portfolio total or an unselected project can no longer appear, and a question about data the selection does not hold says so plainly instead of answering for the whole book.
  • FixedAn Arabic question now always gets an Egyptian-Arabic answer (detected from the text), under the same project restriction, and every requested field is answered — with anything unavailable named explicitly rather than substituted.
  • FixedCurrency table: small rates no longer round to “0.00” — precision widens to keep the significant digits.
  • FixedCSV analysis no longer asks for a literal “A header row” or “One row per record” column — those are structural expectations, not columns.
v1.26

Occupational safety and environmental codes added to the catalog.

  • NewStandards & Codes now covers worker/site safety — Labor Law 12/2003 (OH&S), Ministerial Decree 211/2003, construction site-safety practice, ISO 45001 and OSHA references — and environmental compliance: Environment Law 4/1994 (amended by 9/2009), the EIA/EEAA regime, and construction dust, noise and waste controls. The Ask knowledge base picks these up automatically.
v1.25

Read any answer aloud, project-scoped answers, and a self-repairing call.

  • NewEvery answer now has a “Read aloud” button — press to hear it in a natural human voice (ElevenLabs when configured, your device’s voice otherwise), press again to stop. It reads Arabic answers in Arabic and English in English.
  • ImprovedAsk answers are strictly grounded in the data and the project in scope: pick a project and the reply is about that project only; on the whole portfolio it spans everything you can see — always cited.
  • FixedThe live call no longer ends the instant it connects: the agent is refreshed automatically so the chosen language takes effect, and if a call does drop it now says why instead of a silent “call ended”.
v1.24

One Talk button for voice — the live call or voice chat, automatically.

  • ImprovedThe separate “Voice chat” and “Call” buttons are now one “Talk” button. It starts a live human-voice call when the voice agent is set up (asking the language first), and otherwise the built-in voice chat — so there is one clear way to talk, with the best available experience. Dictate (voice-to-text) stays its own button.
  • ImprovedThe live call now sets itself up on first use for an owner — no separate agent-creation step — when the ElevenLabs key has Conversational AI access.
v1.23

Choose the call language, and a bilingual construction expert on the line.

  • NewPressing “Call” now asks the language first — Egyptian Arabic or American English — and connects the live agent in that language (you can still speak either way once connected).
  • ImprovedThe voice agent is one bilingual construction consultant: whichever language you use, it answers as a professional with the right quantity-surveying and construction terms (مقايسة، مستخلص، غرامة تأخير، المسار الحرج …) rather than generic replies.
v1.22

The voice call fixes itself when the agent is stale.

  • FixedPressing “Call” no longer dead-ends on “ElevenLabs returned 401”: when the stored agent was created with a different or older key, the platform now re-creates it with the current key and connects. If the key itself lacks Conversational AI access, the message says so instead of a bare 401.
v1.21

Voice-agent setup recovers from a stale agent, and the admin log paginates.

  • FixedSetting up the voice call no longer fails with “ElevenLabs returned 401 updating the agent”: if the stored agent can’t be updated (deleted, from another account, or the key was rotated) the platform now creates a fresh agent, and a real key problem is reported clearly.
  • ImprovedThe admin Activity log is now paginated (15 per page, newest first) so it never becomes one very long page.
v1.20

A more human voice, and a conversation that talks back.

  • ImprovedVoice chat now replies to greetings and “can you hear me?” like a person on a call, in English or Egyptian Arabic, instead of “I don’t have that in the project data”.
  • ImprovedWhen ElevenLabs is not configured, the built-in voice now picks the most natural voice your device has and speaks more warmly — and a provider key (ANTHROPIC/OPENAI/ELEVENLABS) can now be set by environment variable, so setting ELEVENLABS_API_KEY switches the voice to a real human ElevenLabs voice and enables the live agent call.
v1.19

Live market news, and a round of QA fixes.

  • NewMarket & News — a new screen with live construction, building-materials and market headlines for Egypt (updated continuously), alongside today’s exchange rates, so the team sees what is moving the market.
  • FixedA selected imported project now shows its own data or a clean “no records yet” on every tab — the Project 360 “Planning/Cost/Safety” tabs, Handover, and the Portfolio, Command Center and Autopilot analytics no longer show another project’s sample figures under the header.
  • FixedPortfolio, Reconciliation and the AI now count the same projects — archived projects are excluded everywhere, so the header, the checks and the AI answer can no longer disagree; a reconciliation check now guards this automatically.
  • FixedImporting a project: a failed validation (for example, completion before start) now clears the old preview and confirmation, and editing any field re-requires a fresh preview — a stale preview can no longer be added.
  • FixedAI providers can now be configured by environment variable (ANTHROPIC_API_KEY, OPENAI_API_KEY, ELEVENLABS_API_KEY), so document analysis, Ask and voice light up on a deployment without needing the console.
  • FixedThe currency converter now rejects non-numeric or negative input with a clear message, and shows small values with adaptive precision instead of “0.00”.
v1.18

A health check for uptime monitoring.

  • NewA public /api/health endpoint reports the running version and whether the database is reachable, and a scheduled check pings it every 15 minutes and raises an alert if the app is down — so an outage is noticed before a client reports it.
v1.17

The notifications bell now remembers, and respects your access.

  • ImprovedNotifications you have read stay read across reloads, and the bell only surfaces alerts for projects and screens your account can actually open.
  • FixedThe notifications panel is now fully translated into Arabic (the empty state and the Command Center link were still in English).
v1.16

Forex now shows a trend, a movement alert and a freshness badge.

  • NewThe Forex screen records the daily rates and draws a trend sparkline and a “moved X%” figure per currency, with a banner flagging any currency that has moved more than 2% — the history builds up over time, and a currency with fewer than two days simply shows “collecting…”.
  • NewA freshness badge shows whether the published rates are live or may be stale, based on the feed’s own timestamp.
v1.15

Search the codes catalog.

  • NewStandards & Codes now has a search box — filter the whole catalog by code, title, scope or authority (for example, “concrete”, “fire”, “HBRC”), with a live match count.
v1.14

Ask can understand a question worded unlike the answer.

  • NewWhen an embeddings provider is configured, Ask now falls back to meaning-based (semantic) retrieval whenever the exact words miss — so “what holds my money back?” still finds retention. With no provider it behaves exactly as before, and it never spends beyond the AI budget.
v1.13

Turn a document’s recommendations into tracked actions.

  • NewAfter the AI analyses an uploaded document, each decision and action now has an “Add to project” button — one click logs it to the project in scope (or the portfolio) as a tracked, auditable item, reversible for 24 hours. The analysis still changes nothing until you choose to apply it.
v1.12

An imported project’s Cost screen now shows its own BOQ.

  • NewSelect an imported project and open Cost & Commercial: instead of “no records yet”, it shows the cost breakdown from that project’s own Bill of Quantities — the value in each package, the largest line items and the line count, clearly labeled as derived from the uploaded BOQ.
  • ImprovedScreens a BOQ cannot speak to — schedule, risk, workforce — still say “no records yet” for an imported project, because a bill does not carry a program or a risk register.
v1.11

Full Arabic across the newer screens, and a hardened quality gate.

  • ImprovedEvery screen added recently — activity, reconciliation, Forex, Standards & Codes, What’s new, voice chat and the live call, document analysis — is now fully translated into professional Egyptian Arabic, with no English falling through.
  • FixedA data-integrity check that had drifted now reconciles the portfolio-exposure decision against the count of at-risk projects, so the number the decision cites always matches the register.
  • ImprovedAutomated tests and continuous integration now run on every change — the type-check, the unit tests, the router eval, the data-integrity checks and a production build must all pass before anything ships.
  • ImprovedThe monthly AI budget is now a hard limit, not just a warning — once the measured spend reaches the cap, paid calls stop (chat and voice fall back to the free built-in answers), and each account has a per-minute request limit so no single login can burn the budget.
v1.10

A full codes catalog, and Ask knows the codes and the platform.

  • NewStandards & Codes now carries a full catalog — the ECP series (concrete, loads, steel, foundations, services, energy), Egyptian Standards, fire and building law, and the international codes referenced in Egypt (ACI, Eurocodes, BS, ASTM, ISO, NFPA, FIDIC).
  • NewThe codes and the platform’s own content are in the Ask knowledge base — ask about any code or how the platform works, by chat or on a voice call, with citations.
  • FixedRemoved the weather widget’s blocked geolocation request — no more console warnings; it uses the Cairo default.
v1.9

A selected project shows only its own data — and more QA fixes.

  • FixedAn imported project no longer shows another project’s sample figures — every module tab, Command Center panel and recovery action shows “No records yet” until it has data.
  • ImprovedA spreadsheet or CSV is now analyzed even without an AI provider configured (columns, data quality, rates vs the Egyptian market).
  • FixedThe report dialog now opens above the tab strip, and reports offer a clear “Print / Save as PDF” and a direct “Download CSV”.
  • FixedReconciliation describes only the checks it performs (portfolio-level ties on the live dataset).
  • FixedThe currency converter flags invalid input and keeps precision for small values.
v1.8

Your imported BOQs become reusable samples, and richer sample BOQs.

  • NewEvery BOQ you import is saved and offered on the Import screen under “Your imported BOQs”, so it can be loaded again in one click.
  • ImprovedThe bundled English and Arabic sample BOQs are now full Egyptian bills — preliminaries, all trades, MEP, medical gases, provisional sums and contingency (EGP 512M and EGP 757M).
v1.7

This page — every update, in one place.

  • NewWhat’s new — a running record of every release, newest first, with a dot in the sidebar until you have read the latest.
v1.6

Standards & codes, live currency rates, and a tour you can re-open any time.

  • NewStandards & Codes tab — the Egyptian construction codes (ECP 203/201/205, Egyptian Standards, Fire Code, Building Law 119) the AI now reasons against and cites.
  • NewTraining library — upload your own contracts and specifications in Data & Insights so the AI reads them alongside the codes.
  • NewForex Rates tab — live online exchange rates for every currency against the Egyptian pound, with a converter for foreign-currency packages.
  • NewGuided tour button — launch a comprehensive, spotlighted walkthrough of every tab at any time, not just on first visit.
v1.5

Staging QA fixes — data integrity, reports, mobile and AI availability.

  • FixedA scoped project no longer shows another project’s sample decisions — it shows “no records here yet” instead.
  • FixedReconciliation now checks the same live workspace dataset the dashboard shows (imported projects included).
  • FixedIndividual report downloads work, and a PDF-labeled report downloads a PDF (with a CSV option).
  • FixedReversed project dates (completion before start) are rejected at import.
  • FixedThe public homepage no longer scrolls sideways on a phone.
  • ImprovedAI document analysis shows whether it is available before you upload; multi-part questions now return every requested figure.
v1.4

Nothing is ever deleted, clearer activity, and fully project-scoped tabs.

  • NewArchive — deleting a project, demo account or presenter moves it to an Archive you can restore from; nothing is hard-deleted.
  • NewPer-user activity with pagination, so the log never becomes one very long page.
  • ImprovedChoosing a project now scopes every module tab — totals allocate to that project, matching the headline.
v1.3

A live voice call with the assistant.

  • NewVoice call — a real spoken conversation with the assistant (ElevenLabs), in Egyptian Arabic or American English, configurable in Admin → AI.
  • ImprovedDeeper project scoping across module tables and lists.
v1.2

Understand any uploaded document.

  • NewData & Insights reads any file — a contract, BOQ, variation order, payment certificate or site photo — and returns the findings, decisions and actions.
v1.1

A construction expert that speaks Egyptian Arabic.

  • ImprovedThe AI answers as an Egyptian construction consultant in Egyptian Arabic or English, in text or voice.
  • NewTwo ready-made sample projects — an English residential tower and a full Arabic hospital BOQ — loadable in one click on the Import screen.
v1.0

Voice chat, read-only presenters, and the pre-demo check.

  • NewHands-free voice chat in Ask ConstructINS, in English or Egyptian Arabic.
  • NewPresenter accounts — read-only, client-like logins with owner-controlled per-project access.
  • ImprovedA working “re-run” on the pre-demo readiness check, with a fix link on each item.

What’s new on the other InsAI platforms? News from every platform →