Production plan OR-Tools
- Optimal: margin 4,068,409.
- Bottling hours in Bergen are at their limit (1,600): one more adds 75 to the margin.
- IPA demand is at its limit (60,000): one more liter adds 19.5.
Alpha · OpenBI 0.3.8 · written in C · made in Norway
Business intelligence for the age of agents.
A BI engine and studio in one install. Your organization's measures live in a semantic model, so any agent — Claude, ChatGPT, Grok or your own — answers with numbers it cannot make up. No Python. No R. Nothing to pip install.
- Claude
- ChatGPT
- Grok
- your own agent — anything that speaks MCP
51
visuals, from ledgers and Sankey diagrams to routes on a map
98
tools for agents over MCP: model, queries, reports, notebooks, versions
2,549
DAX queries checked against Power BI's own answers. 0 failures.
1
signed installer. No Python, no R, no packages to keep alive.
Why another BI tool
Python and R were made for humans.
They were built for people at a keyboard, one package at a time — and they are wonderful at that. But an agent doesn't need a REPL and forty dependencies. It needs one engine that answers in milliseconds, and says no when a name doesn't exist.
OpenBI's engine is written in C. DAX, SQL and Power Query run inside it, and so do the libraries that do the heavy lifting — DuckDB, OR-Tools, QuantLib, igraph, LightGBM and Typst. One engine, one model, one answer.
“The fastest dependency is the one you never installed.”
a stack of packagesone engine
Your wingman
Talk to me, Goose.
You're never alone. DAX, M, MDX, SQL — the syntax you used to fight is a thing of the past. Ask in words; your agent writes the query, OpenBI checks it against the model, and the answer comes back in words.
Recorded from a real session over MCP:
ClaudeChatGPTGrokyour own
OpenBI Studio — the DAX viewConnect any agent in one line (the studio shows yours, with its own key):
claude mcp add --transport http openbi-studio http://127.0.0.1:5739/mcp \
--header "Authorization: Bearer <your studio's key>"
Other agents take the same address as JSON. The studio says when one is connected, and how many calls it made.
Bond portfolio QuantLib
- Market value 884.94 M; a one-basis-point rise in rates costs 318,623.
- Rates +100 bp: −30.98 M; convexity softens the DV01's estimate by 884,689.
- VaR 99 %, one day: 3.79 M.
Who buys what igraph
- 19 nodes, 60 edges, 2 communities (modularity 0.37: a clear structure).
- Most central: Pils, IPA, Cider.
- No cut points: no single shop holds the network together.
The agent's explanations, word for word, from the studio's explain tool on the example project.
The semantic model
An analytical language that can't be hallucinated.
Like SQL, Open Analytics Expression (OAX) is exact: a query names measures and columns that exist, or it fails. Your organization's measures, relationships, roles and calendars live in one semantic model — written once, by people who know what “revenue” means — and every agent uses the same definitions.
Measures with descriptions
Relationships
Roles and row-level security
Calendars and time intelligence
TMDL text, in git
measure 'Revenue' = SUMX(Sales, Sales[Liters] * RELATED(Products[Price per liter]))
formatString: #,0
/// NOK of revenue for each liter sold
measure 'Revenue per liter' = DIVIDE([Revenue], [Liters sold])
formatString: #,0.00
A replay of the example project's answers, recorded from OpenBI. The misspelt names are refused exactly as the engine refuses them.
Versions
Made a mistake? Recover in a breeze.
Every change — yours or the agent's — is tracked in git and saved as versions you name. The model is text, so a change reads like a sentence, and going back is one click.
- e3968e4The brewery as it came
- 085341dAdd Revenue per liter: NOK of revenue for each liter sold[Revenue per liter] = 44.36
- a mistake[Revenue] deleted[Revenue per liter]: the value for column 'Revenue' cannot be determined
- back to 085341dRestored — the model loads again as saved[Revenue per liter] = 44.36 · 0.6 ms
The real sequence, from the example project — the agent made the mistake on purpose.
Trust, but log
Don't trust agents? Neither do we.
- Every step an agent takes is in the activity log, with its badge and the time: the queries it ran, the measures it changed, the versions it saved.
- It never sees a password or a key: when a source needs signing in, the studio asks you, and the agent waits.
- Roles and row-level security in the model decide what each reader sees — and “see as” shows you exactly that.
- It is told to save a version only when you agree — and any version can be undone.
we detect everything
The visuals
51 visuals. One runtime.
Every picture below is OpenBI drawing the example project — captured, not mocked up. Flip the owl switch and they all turn dark, because the visuals do too.
Chart themes
Do you love ggplot? We've got you covered.
Or something more serious? Nine looks, one attribute on the page — the same chart, the same data, drawn by OpenBI in each.
Are you an owl?
We have a dark theme — for the studio, the reports and this page.
Financial reporting
Statements as finance prints them.
A chart of accounts read as a parent-child tree, added up by its operators. Actual against budget and last year, variance bars, comments on figures — and out the other end, Excel with live formulas, or a typeset PDF.
The same statement as a typeset PDF · move to zoomExcel with formulas
Totals and variances as formulas, the rows as an outline, the figures exact with the scale in their number format.
Typeset PDF
Typst on the server: rules, a double rule under the total, headings on every page, comments as footnotes. PDF/A on request.
Comments on figures
Right-click a figure to comment. Kept with the report, who and when, under the selection they were written for.
Formulas
Are you a mathematician? LaTeX is there for you.
Write formulas in TeX on any report page, and put the model's live values in them with \val{[Measure]} — they follow the page's filters. Below is the same renderer OpenBI ships. Type away.
One tool
Everything inside. Nothing to install.
Operations research, quantitative finance, network analysis and prediction — next to OAX, SQL and Open Query, in the same engine, on the same model. No second runtime, no extra packages.
OR-Tools
Operations research
Production plans, vehicle routes, schedules — with what limits them and what one more of each is worth.
igraph
Networks
Communities, centrality, bridges — who matters, and why.
QuantLib
Quantitative finance
Curves, bonds, options, FX, VaR — and QUANT.* functions right in DAX.
notebooks
Code first
DAX, SQL and C# steps, versioned and approved by a person before they reach the model.
LightGBM
Prediction
Churn, propensity, forecasts — trained on the model's history, scored with PREDICT() in DAX.
DAX enginePower Query M engineSQL · DuckDBOR-ToolsQuantLibigraphLightGBMTypst PDFsXMLA for Excel
Notebooks
Every question, a notebook. No compute bill.
Forecasts with LightGBM, valuations with QuantLib, plans and routes with OR-Tools, networks with igraph — each a notebook of OAX, SQL and C# steps that runs on your own computer, beside the model. No cluster to rent, no cores billed by the hour, and your data stays where it is.
LightGBMHow much will we sell in the next 6 months? Strong: off by 2 % over the six months held back — a simple rule was off by 3 %. A person approves it before it forecasts in the model.
QuantLibWhat are the treasury's bonds worth, and what drives their risk? Market value 884.94 M, DV01 −318,623: a basis point costs 318,623. VaR 99 %, one day: 3.79 M.
OR-ToolsWhich shops does each of three vans visit, and in what order? Three vans, sixteen shops, 65.7 km in all; the longest round is 26.9 km.
igraphWhich shops buy alike, and which products tie them together? Nineteen nodes, sixty edges, two communities; Pils, IPA and Cider matter most.
0 kr
in compute
It runs on the cores you already have — your laptop, or a server of your own. Nothing to rent, nothing metered.
Approved by a person
The agent saves drafts and never approves. Someone other than a version's author approves it before it is used.
All in one place
The model, its notebooks, their versions and the reports: one project folder, one git history.
Security
Your fast track through the security review.
Fewer moving parts, fewer surprises. No interpreter to patch, no package manager pulling code at run time — one installer, signed, with its bill of materials inside.
0known vulnerabilities
in the 45 components that ship, checked against OSV and NVD when the installer is built — a build with one stops.
1signed installer
OpenBI's executables and engine signed by Norway Analytics; a CycloneDX bill of materials installed with the studio.
0secrets for agents
Credentials stay with the studio. PDF typesetting runs in an empty folder that can read no file and fetch no package.
Excel
Old-fashioned and love Excel? Darling, we've got you too.
Click Analyze in Excel and a PivotTable opens, live on the model: Excel speaks to OpenBI the way it speaks to Analysis Services. Your CUBE formula dashboards can come right back.
- PivotTables and slicers over the live model — the measures and roles you defined.
- CUBE formulas: the MDX they send, answered by OpenBI.
- Statements to .xlsx with their formulas, when you'd rather have a file.
Book1.xlsx — connected to OpenBI · Example Brewery
C3=CUBEVALUE("OpenBI","[Measures].[Revenue]","[Products].[Product].&[Pils]")
| A | B | C | |
|---|---|---|---|
| 1 | Example Brewery | NOK / liters | |
| 2 | |||
| 3 | Revenue | Pils | 38,164,388 |
| 4 | Liters sold · Frogner | Pils | 115,604 |
| 5 | Liters sold · Frogner | IPA | 3,674 |
| 6 |
The figures are the model's answers to those queries, through MSOLAP — the provider Excel uses.
Coming from Power BI
One of Power BI's 30 million users? Come row with the Norwegian team.
Ro!Norwegian for “row” — together, in time.
- Import from Power BI — one click.A .pbip project, a semantic model folder or TMDL becomes an OpenBI project.
- Everything comes along.Tables, measures, relationships, roles, calendars — and Power Query, which OpenBI runs in its own M engine.
- Your DAX answers the same.2,549 queries recorded from Power BI's engine: 2,318 identical, 220 refused by both, 11 known and documented differences.
- Your agent rebuilds the pages.Report pages are HTML in your project, so the agent writes them — and you review them in git.
30 million monthly users: Microsoft's own figure, from its 2025 recap. Power BI is a trademark of Microsoft; OpenBI is not affiliated with Microsoft.
Enterprise partner
Bringing OpenBI to a whole organization? Meet Querex.
Querex AS is OpenBI's Enterprise Partner. They connect an organization's databases, semantic models, files and APIs into one workflow — with the access you authorize — and put agents to work on it: from data to decisions in seconds.
Visit querex.no
Enterprise Partner
Download
In alpha. Get a seat in the boat.
OpenBI is in alpha and changes fast — often from one week to the next. Tell us who you are, show us you're not a robot, and the installer is yours.
- It changes every week. New versions and new features — and now and then, something that breaks.
- Your work stays safe. Projects are plain files in git: any version comes back with one click.
- You help steer. The first rowers' feedback sets the course.
- Teams welcome. Installing for a team? Say so in the message.
Your download has started.
If it hasn't, download the installer. The link works for 24 hours. When you run it, Windows names the publisher: Norway Analytics. Ro!