Open-source architecture governance
Stop running architecture governance on slide decks and spreadsheets.
Review boards that meet monthly over PowerPoint. Decisions buried in minutes and inboxes. Approval conditions nobody follows up. And when an auditor, a regulator or a new CIO asks what was approved, why, and whether the estate still matches it, the answer takes weeks of digging. Archigov turns all of that into one system of record: decisions made in days, on the record, and checked continuously against what you actually run.
docker compose up --build # one container, one SQLite file — usable in minutes
Decisions linked to the frameworks you answer to
Sources: SAP LeanIX EA Insights 2025 (n=360); vFunction architecture survey 2025 (n=600+). Vendor-sponsored surveys, quoted as published.
The Problems We Solve
Architecture governance rarely fails for lack of process. It fails because the process lives in tools that cannot hold it: decks, wikis, inboxes and spreadsheets.
Decisions take weeks
Designs wait for a monthly architecture review board, get rebuilt as decks "for the board", and come back a third time with nothing to show for it.
With Archigov Reviews are risk-tiered. Area owners record positions when they are ready, low-risk designs that follow your patterns clear in a fast lane, and a sitting is only for what needs one.
The record is a slide deck
Rationale lives in decks, minutes and email. Six months later nobody can say what was decided, why, or who agreed.
With Archigov Every architecture decision record (ADR) carries its argument, the options weighed and the risks accepted. Each position is recorded by its owner, and the minutes are derived from the record.
Conditions are forgotten
Approvals come "with conditions" and exceptions "for now". Nobody tracks them, and waivers never expire.
With Archigov Conditions and exceptions become obligations with an owner and a date. Overdue ones land in the owner's queue and in the board's metrics.
Approved isn't what's built
Once a design is approved, nothing checks the estate against it. Drift is found in the next incident or the next audit.
With Archigov Decisions carry checkable constraints that run continuously against the estate collected from Azure. Drift raises a finding and goes back to the board.
Audits become archaeology
DORA, ISO 27001, the EU AI Act and clinical-safety standards ask for a traceable chain from risk to decision to change. Every time, it is rebuilt by hand from SharePoint and inboxes.
With Archigov The chain is the record. A signed evidence pack and regulator-ready exports take one click. They show evidence and who stands behind it, never a claim of compliance.
Repositories go stale
EA repositories are populated by hand, priced per application or per seat, and used by a handful of architects. Everyone else never logs in.
With Archigov The model is derived from your cloud estate in minutes, not drawn. Pricing is flat per instance with unlimited users, so everyone who decides takes part.
What Changes for Each Role
Governance that is faster because it is structured, with one record that serves everyone who answers for the architecture.
Run your review board in days, not months
No decks to assemble, every decision on record, and the board's lead times, recycling and backlog measured from the records.
Know what was approved, what was built, and where they differ
A roadmap built from decisions, and how far the estate has realised each one, without anyone drawing it.
An auditor-ready trail, generated as you work
Risk to decision to approval to change, with conditions, exceptions and accepted risks. Signed, chained and exportable to OSCAL and the DORA register.
Guardrails, not gates
Designs that follow your golden-path patterns and pass every check clear without waiting for a sitting. The board reviews only the exceptions.
How Archigov Compares
Most teams already govern architecture with something. Here is where each approach breaks, and what Archigov adds.
| What you use today | Good at | Where governance breaks | What Archigov adds |
|---|---|---|---|
| Slide decks, wikis, tickets and email | Flexible, familiar, already paid for | No approval record or sign-off evidence; conditions and exceptions untracked; nothing links a decision to what is deployed | The same reviews as a structured workflow with a permanent, signed record, and no per-seat licence for the people who take part |
| Decision records in Git | Decisions next to the code, under version control | No board, no approval workflow, no audit trail, and no check that the estate still matches the decision | Keeps the familiar decision-record format and adds who decided, under what conditions, and whether it still holds |
| EA repository suites | Portfolio and capability modelling, rich analytics | Populated and maintained by hand; priced per application or seat; review boards still run on decks beside them | Governance first, with the model derived automatically and flat pricing. Works alongside a suite through ArchiMate import and export |
| Cloud security posture tools | Checking resource configuration against controls | They check resources, not the design decisions behind them, so they cannot say why the architecture is the way it is | Imports their results (Defender for Cloud, Azure Policy) as evidence and ties them to the standards and decisions they prove |
How It Works
One record follows each decision from the first draft to the day it is superseded: argued properly, decided by the right people, and kept true afterwards.
- 01 · ProposeRecommended option, justification and consequences — with the risks accepted stated. AI tests the argument.
- 02 · LinkEvery decision cites the standards, patterns, policies and earlier decisions it rests on.
- 03 · DecideThe right board decides: area owners record positions, the chair declares, conditions are tracked.
- 04 · EnforceA checkable constraint runs against your real estate; drift raises a finding and returns to the board.
- 05 · EvolveAccepted, conditions met, superseded with lineage — every step signed and chained.
AI You Can Audit
AI that drafts decisions nobody can verify is a new risk, not a fix. Archigov's eleven optional AI features test arguments and pre-review designs, but every citation must resolve to a real standard, decision or resource, and every output is marked AI-generated. Nothing an AI writes approves, blocks or becomes the record without a named person's audited act, and every feature has a deterministic fallback.
- Decision review — a critical test of the argument before it reaches the board
- Pre-review — cited findings on a design; citations must resolve or they are discarded
- Rule drafting — natural language to a checkable rule, validated and dry-run first
- Evidence suggestions — missing evidence becomes a tracked assessment in one click
Draft: “Expose the Referral API directly to the internet with IP allow-listing — effectively as secure as a gateway.”
Open Source, Without the Rug-Pull
The core is AGPL-3.0. Run it yourself — even modified — for your own organisation with no obligations. Self-host from one container with SQLite inside; continuous Litestream replication if you want it. A managed instance, hosted in the region you choose, is available for teams that would rather not run it: hello@archigov.com.
Get the codeThe commitment, made on day one: this project will never move to a licence more restrictive than AGPL-3.0. No BSL, no SSPL. Policy packs, assessment templates, connectors and importers are Apache-2.0, so sharing them, or integrating with Archigov, never requires touching AGPL code. The commercial licence covers the same code and holds nothing back from the open-source release.
Pricing
Flat per instance. Unlimited users — governance only works when everyone can take part, so there is no per-seat pricing, ever.
Self-Host
£0
AGPL-3.0, forever
- The whole product — nothing held back
- Unlimited users and boards
- One container, your infrastructure
- Community support on GitHub
AI
- Bring your own keys — integrations for OpenAI, Claude, Mistral, Ollama and any OpenAI-compatible endpoint
- You pay your provider directly; no AI fees from us
Supported Self-Host
from £3,000
per instance, per year
- Everything in Self-Host
- Vendor support with response times
- Upgrade and migration assistance
- Scales with estate size
AI
- Bring your own keys — integrations for OpenAI, Claude, Mistral, Ollama and any OpenAI-compatible endpoint
- You pay your provider directly; no AI fees from us
Managed Instance
from £20,000
per instance, per year
- We run it; you govern
- Dedicated instance, continuously replicated
- Hosted in the region you choose: Western or Eastern Europe, Eastern or Western North America, or Asia-Pacific
- EU or US jurisdiction available: the instance's data is stored and processed only there
- Your Entra ID, your evidence keys
AI
- Included: hosted models with no keys to manage. Inference runs in the EU, whichever region hosts your instance
- AI Premium add-on: frontier models under zero-retention contracts, from £3,600/yr
- Additional usage from £0.20 per action
- Need AI inside your own region? Use your own model tenant (e.g. Azure OpenAI in your region): £2,000/yr connector
Compare Options
| Detail | Self-Host | Supported Self-Host | Managed Instance |
|---|---|---|---|
| Product and licence | |||
| Price | £0 | from £3,000 / year | from £20,000 / year |
| Licence | AGPL-3.0 | AGPL-3.0 | AGPL-3.0 (we operate it) |
| Features | All of them | All of them | All of them |
| Users and boards | Unlimited | Unlimited | Unlimited |
| Hosting and operations | |||
| Who runs it | You | You | We do |
| Where it runs | Your infrastructure | Your infrastructure | Dedicated instance in your chosen region: Western or Eastern Europe, Eastern or Western North America, Asia-Pacific; EU or US jurisdiction available |
| Backups and replication | You (Litestream example included) | You, with our guidance | Continuous replication, run by us |
| Upgrades | You | Assisted upgrades and migrations | Done by us |
| Sign-in | Local or your Entra ID | Local or your Entra ID | Your Entra ID |
| Evidence-signing keys | Generated on your instance | Generated on your instance | Per instance; public key published to you |
| Support | |||
| Support channel | Community (GitHub) | Vendor support | Vendor support |
| Response times | Best effort | Agreed response times | Agreed response times |
| Procurement paperwork | — | Standard | Standard, incl. data processing terms |
| AI | |||
| AI provided by us | No | No | Yes — included |
| Provider and keys | Yours: OpenAI, Claude, Mistral, Ollama or any OpenAI-compatible endpoint | Yours: same integrations | Ours: hosted models, no keys to manage. Inference runs in the EU, whichever region hosts your instance |
| AI fees from us | None — you pay your provider | None — you pay your provider | Included allowance; additional usage from £0.20 per action |
| Frontier models | With your own keys | With your own keys | AI Premium add-on, from £3,600 / year |
| Your own model tenant | Yes (your keys) | Yes (your keys) | £2,000 / year connector: AI stays in your tenant and the region you deploy it in |
Need UK-only data residency? Run Self-Host or Supported Self-Host in your own UK tenant: it is the same single container. Prices are per instance and indicative; estates differ, so write to hello@archigov.com and we will quote plainly.
Make every architecture decision count — and keep it true.
Start with the demo in five minutes: propose a decision, watch the AI test the argument, take it to the board. Or talk to us about a managed instance in your region.