AI review
Review a design doc against the code before you build it
Pick a model, and ArchDocs has it read the whole design doc, your organization's other docs and decisions, and the code in the repositories the doc links to. It records findings with evidence. People decide what to do with them.
Plans: AI review uses AI credits from your organization's shared balance, so it is available on Team, Business and Enterprise. A higher-rate model uses more credits for the same review.
Illustration of an AI review of DD-0031 with two open findings. The first, a high-severity premise finding, says the consumer list comes from gateway logs and may miss callers outside the gateway.
Reads
It reads the doc, the context and the code
A review starts from the whole design doc: every variation, the open questions, the work log and the comment threads. It can search your organization's other design docs, ADRs, objects and diagrams, and it only sees records visible to the whole organization.
- Code from the repositories the doc links to, through code locations, linked pull requests or catalog objects, each pinned to a commit when the review starts
- A short-lived sandbox per review, kept apart from the servers that run ArchDocs; the reviewer can search and read files but can't run your code
- Models from Anthropic, OpenAI and Google, limited to the providers and models your admins allow
Findings
Findings with evidence you can check
The reviewer tests the doc's premise and looks for a better approach before it checks the details. Each finding has a kind, a severity and its evidence. Code evidence links to the exact lines at the pinned commit.
- Premise: an assumption behind the design is wrong, unproven or out of date
- Better approach: another solution would work better, with its trade-offs
- Risk: the design could fail or break something in a specific way
- Gap: something the change needs is missing from the doc
- Inconsistency: the doc contradicts itself, the code, an ADR or another doc
- Post a finding as a comment anchored to the passage it is about, discuss it, resolve it or dismiss it
Conversation
Keep talking to the reviewer
A review keeps its context, so you can ask why it raised something, push back, or come back days later. When the doc changes, Re-review reads it again, moves each repository to the latest commit on its branch, adds new findings and marks earlier ones it now sees as fixed.
Control
Admins decide what it can reach
- Turn each provider on or off and choose the allowed models
- Code access: linked repositories only, or linked repositories plus ones members add
- Web research on or off
- Automatic reviews when a review is requested or a variation is marked ready, off by default
- Agents can read reviews over MCP; starting one over MCP is off by default
Limits
Advice, not approval
An AI review never approves a variation, marks one ready or counts toward required approvals. Its findings are a model's reading of the doc and the code, so check the evidence before you act on one.
Plans: AI review uses AI credits from your organization's shared balance, so it is available on Team, Business and Enterprise. A higher-rate model uses more credits for the same review.
Screenshots are from a sample workspace. Recreated views are labeled.