Five stages, each standing on the one before
We didn't set out to build a suite. We set out to fix how organisations work with their data, and each problem we solved exposed the next one.
- Chapter 01 · Own your dataShipped · Open source
Data OS
One engineer-native workspace to ingest, transform and deliver data.
Data work is scattered across too many tools. Integration lives in one product, transformation in another, reporting in a third, and the glue between them is scripts nobody owns. Every handoff adds licence cost, latency and a new place for things to break.
- Polyglot notebooksSQL, Python, JavaScript and HTML cells side by side in a single, versionable YAML document.
- Embedded analytical engineDuckDB in-process as the shared data layer, so there's no database server to stand up.
The foundation every other stage builds on.
regional-sales.flownbsql✓CREATE TABLE sales AS SELECT region, SUM(amount) AS revenue FROM dataverse.opportunities WHERE statecode = 'Won' GROUP BY region;
python✓df = data.get("sales") df["share"] = df.revenue / df.revenue.sum() data.put("sales_share", df)regionrevenueshareGCC4.2M46%Levant2.9M32%Europe2.0M22%html · vega-lite✓4.2MGCC2.9MLevant2.0MEuropehtml · report✓regional-sales-report.pdf3 pages · rendered from the sales_share tableRan 4 cells · zero-copy Parquet hand-off · chart and report rendered inline
- Chapter 02 · Run it like infrastructureIn build · Design-partner access
Enterprise Operations
Take the same workflows to production with governance, audit and separate environments.
A notebook on a laptop isn't a production service. Production means knowing who can run what, where secrets live, how a change moves from dev to prod, what happens when it fails, and being able to prove all of it to an auditor.
- Identity & accessLocal accounts plus SSO via Microsoft Entra ID, Google and generic OIDC, with SAML and LDAP planned. Roles are scoped per environment and per workflow.
- Environments & promotionDev, QA, test, prod, or as many as you need. Each has its own setup, and changes reach it only through a governed pipeline.
Runs the Data OS engine as-is. It extends the core, it doesn't modify it.
Promotion · regional-salesdevv14Deployedqav13Verifiedprodv12Awaiting approvalPlan · qa → prod2 approvals required- + connection dataverse → prod-crm (mapped)
- + secret crm-client → Key Vault
- ~ parameter region_filter changed
Audit chaina41f9c02e7b351d80fa6 - Chapter 04 · Automate the business processComing soon
AI for Business
Agentic AI that runs your business processes end to end, not just the data behind them.
Process automation breaks at the first exception. Rule-based workflow tools handle the happy path, and everything else lands back in someone's inbox. Most processes are largely routine with a layer of judgement on top, and that judgement is where the time goes.
- Agentic process flowsMulti-step processes where agents plan and carry out each step, rather than single prompt-and-response exchanges.
- Human approval where it mattersApproval gates on anything irreversible or high-value, with the agent's reasoning shown alongside.
Uses the Data OS for data, Enterprise Operations for governance, and the grounded-agent approach of AI Data Authoring.
Process run · invoice-to-pay- TriggerSupplier invoice received by email
- AgentExtracted 14 line items, matched PO-4471 in Dynamics 365
- AgentPrice variance of 6% exceeds the 5% policy, so escalating
- HumanFinance approver signed off with a note
- AgentPosted invoice, scheduled payment, notified supplier
5 steps · 1 human decisionEvery step auditedIllustrative example. This stage is in design.
- Chapter 05 · Architect the whole changeEarly access
Transformation Squad
An AI architecture squad that documents where you are, designs where you're going, and plans the route.
Transformation programmes start blind. The as-is is undocumented, the to-be lives in slides, and the gap between them is discovered expensively, mid-delivery. Architecture work is slow to produce and quickly goes stale.
- Specialist architect agentsBusiness analyst, data, cloud, CRM and principal architect, each owning its own part of the model.
- Enterprise architecture modelElements, relationships, snapshots and a gap analysis, so the as-is and to-be are data rather than diagrams.
Recommends the suite wherever it fits, which brings the story back to the Data OS.
Architecture · as-is → to-beAs-isTo-be3 integration toolsgapOne Data OSManual monthly reportsgapScheduled workflowsSecrets in config filesgapCentral Key VaultDynamics 365 CRMkeepDynamics 365 CRMRoadmapPhase 1 · FoundationPhase 2 · MigratePhase 3 · Automate - Chapter 06 · What comes nextComing soon
The next chapter is already being written.
Data OSBranching & checkpointsConditional cells and checkpoint/rollback in the headless runner.
Enterprise OperationsEnterprise sign-inSAML and LDAP, alongside Entra ID, Google and OIDC.
Enterprise OperationsProduction on AKSAzure Kubernetes Service as the first production target.
AI for BusinessAgentic process automationAgents that run multi-step business processes, with human approvals and a full audit trail.