Sign inCreate free account

How it works

One graph from need to evidence

OpenV keeps a product’s requirements, the reasons behind them, the design that meets them and the tests that prove it in one typed graph, with AI agents that work inside the same audit trail as the people.

The artifact graph: needs, requirements, design items, test cases and test runs, joined by typed linksUser needNEED-6RequirementREQ-16 · v5 · approvedDesign itemDES-7Test caseTC-1Test runpass · evidencederives fromsatisfied byverified byresultChange REQ-16 and every link from it goes suspect until a reviewer confirms it.
Every artifact has a stable reference and a version; every link has a type. Coverage, gaps and impact are read off this graph, not maintained by hand.
The agent loop: an agent reads the project through typed tools, proposes changes, and a person approves them into the projectYour projectartifacts · links · runsAgent runyour own AI, your keysProposala diff, waiting for reviewA personapprove · rejectreads (typed tools)writes back asreviewed byapproved changes land, versioned
By default an agent does not write to the project directly: every change it makes is a proposal with a diff, approved or rejected by a member. A workspace can let a trusted agent write directly instead, and either way each change is recorded like any other version.

From a blank project to a signed-off specification

1. Capture what the product must do

Write requirements by hand, run a guided definition with the V&V Assistant, or send stakeholders an interview link and let an interviewer agent file candidate needs as drafts. Every requirement gets a stable reference on creation and a quality score against the wording convention you chose.

2. Link it to why, and to how

Typed links join needs to requirements, requirements to design items and test cases, hazards to mitigations, and a child project’s requirements to the parent’s they refine. Change the wording and the links go suspect until somebody confirms them.

3. Prove it

Test runs record a result per test case, with evidence attached. Coverage, gaps and the traceability matrix update as links change; verification rolls up from subsystems to the system.

4. Let agents do the legwork, and approve their work

Agents draft test cases, check consistency, find gaps and answer questions, on your own AI subscription. They write back as proposals; a person approves each one, and the approval is in the history.

5. Freeze it and hand it over

Capture a baseline before a review, a release or a contract. Produce the PDF or Word document with exactly the sections and fields you need, share a read-only link, or export ReqIF for the tool on the other side.

Where things run

The hosted service and a self-hosted OpenV are the same stack. Agents run wherever you put a runner: the Agent Connector on your own machine, or a leased cloud runner.

Where things run: the browser talks to the API, the API to Postgres and files; agents run on a runner you control or leaseBrowserweb app · phoneOpenV APIGo · authz on every callPostgresrows per workspaceFilesattachments · figuresRunneryour machine or a leaseAI provider you choseClaude · Codex · Gemini · own keysHTTPStyped toolspromptsYour requirements go to the AI provider you chose, and nowhere else.
The hosted service is one API and one database per deployment, with every request checked against your workspace and project role. Self-hosting is the same stack, on your hardware.
Three levels of access: a public link to read, a reviewer link to comment, a membership to editYour projectPublic linkread, no accountReviewer linksign in, comment, never editMembershipeditor · owner, by namecustomer · assessorreviewercontributor
Read and comment access travel as links because forwarding them is harmless. The right to change the specification is granted person by person.

See it working

Five narrated recordings on the platform’s own requirements project, three on a desktop and two on a phone.