Ship it with proof.

ArcRelease reads the tools you already run and judges every release on the evidence. Shipping past a failure takes a signature. It never sits in your deploy path.

How it judges

Preflight

Example release · illustrative data

4 pass · 1 warn · 1 fail · 1 not checked

  1. Release ceremonies2 of 12 still pendingFrom Ceremony owners
    Warn
  2. Risk score34, under the 60 thresholdFrom Gaja risk model
    Pass
  3. Open incidentsNone correlated to this releaseFrom PagerDuty
    Pass
  4. Change freezeNo active freeze windowFrom Freeze windows
    Pass
  5. Upstream dependenciesEvery dependency has shippedFrom Release graph
    Pass
  6. Approvals1 approval still outstandingFrom Approvers
    Fail
  7. Security findingsNo scan captured for this releaseFrom Snyk
    Not checked
Part of the Google for Startups cohort at T-Hub.Google for StartupsT-Hub

Judged on evidence. Signed when it isn’t clean. Live today

Your tools report

Evidence arrives from what you already connected. No new agent in your pipeline.

Connected today: GitHub, GitLab, Bitbucket, Jenkins, CircleCI, Argo CD, Snyk, SonarQube, PagerDuty, Datadog, Grafana, Jira, Linear, ServiceNow and Freshservice.

Every check gets a verdict

Pass
The evidence says it’s clean.
Warn
Worth a look before you ship.
Fail
Blocks the record until someone signs.
Not checked
No data, no opinion. Never counted as a pass.

Proceeding is on the record

An override needs a statement and evidence. The preflight is frozen at that moment, in a log nobody can edit.

release.risk_acceptedstatement · 1 evidence link · failing checks: approvalspreflight snapshotted · immutable

When it breaks, read it back to the decision. Next

  1. Incident
  2. Deployment
  3. Release
  4. Pull request
  5. Commit
  6. Ticket
  7. Requirement
  8. Review
  9. Ruling

What caused it, who decided, what was checked, what should have caught it. If nothing was checking, that gap becomes the next check. Requirement, review and ruling exist only for changes built through ArcCode.

The front door: start the change here too. Next

ArcCode

Bring the ticket from your tracker, whether Jira, Linear, GitHub or Asana, and the coding tool your team already uses.

  • A spec a human always approves
  • Coverage checked before any code
  • Code review, by a second tool if you choose
  • Every call the agent made alone, acknowledged

Project setup

Set a project up once, and every change inherits it.

  • PRD, FRD and BRD, every version kept
  • Your branching strategy and hotfix path
  • Which stages need a human, per project
  • The SOP your release must follow

Questions teams ask first

Does ArcRelease sit in our deploy path?

No. Your pipeline deploys. ArcRelease reads evidence and records decisions, and if it is down, nothing stops.

What does “Not checked” mean?

No evidence was captured for that check, so ArcRelease gives no opinion. It is never counted as a pass, and it tells you which tool to connect.

How is the risk score calculated?

A deterministic model adds points for evidence such as failed deploys, open incidents, pending ceremonies, security findings and critical dependent services, each within a fixed cap. Every point is listed with its reason.

What happens if we ship past a failing check?

An admin or release manager signs a statement and attaches at least one evidence link. The preflight is snapshotted at that moment and written to an audit log the database will not let anyone change.

Which tools does ArcRelease read?

GitHub, GitLab, Bitbucket, Jenkins, CircleCI, Argo CD, Snyk, SonarQube, PagerDuty, Datadog, Grafana, Jira, Linear, ServiceNow and Freshservice, with notifications to Slack. ServiceNow change requests are captured today; evaluating them as a gate is next.

Which coding agents will ArcCode support?

ArcCode is next, not live yet. It is designed around the agent your team already uses, such as Cursor, Claude Code or Codex, chosen per stage.

How do we start?

With a 30-day, read-only pilot: connect the tools you already run and see your next release judged.

Start with a 30-day, read-only pilot.

Connect the tools you run today and see your next release judged. Nothing to install in your pipeline.