Skip to content

AI AEGIS · AI-powered release tollgate validator

Ship on evidence,
not on gut feel.

AI AEGIS checks requirements, test coverage, API test results, pull request risk and CI status. An orchestrator combines that evidence to recommend GO, CONDITIONAL GO or NO-GO, explains the decision and flags any conditions for human sign-off.

  • No sign-in for the demo
  • Every finding cites its evidence
  • Human sign-off built in

Release run · sample

Checkout: saved cards

Agents live
  • Requirements6 acceptance criteria parsedPASS
  • Test architect14 scenarios mapped to requirementsPASS
  • API testsStaging run · 2 slow endpointsWARN
  • PR riskChanged lines only · score 41 / 100WARN
  • CI checksRead from the pull request linkGREEN

Verdict

0 blockers · 2 conditions

CONDITIONAL GO

Release once the slow endpoints are tuned and PR risk is reviewed. Awaiting sign-off.

Illustrative sample. Run it yourself in the demo

Watch · 1:39

See one story go from Jira to a signed-off release.

A walkthrough of the public demo with sample data. Prefer to click through yourself? Try the live demo

Connects to the tools your release already touches

  • GitHub checks
  • Jira Cloud
  • OpenAPI 3
  • Auth0 · OIDC
  • OpenTelemetry
  • DeepEval
  • PostgreSQL + pgvector
  • Temporal
  • FastAPI
  • Next.js
  • PDF · Markdown · JSON reports
stage agents check every story
4stage agents check every story
possible calls: GO, CONDITIONAL, NO-GO
3possible calls: GO, CONDITIONAL, NO-GO
specialised agents, from requirements to CI repair
6specialised agents, from requirements to CI repair
human signs off before anything ships
1human signs off before anything ships

These describe how the product works, not customer results.

The platform

Every stage of a release, checked by an agent that shows its work.

Every agent below is live today. Plug in Claude for deeper AI reviews whenever you want a second opinion.

How it works

Four steps from a pull request to a decision you can defend.

  1. 01

    Bring the evidence

    A requirement or Jira story, an OpenAPI document, and a pull request diff or link. Paste them or import them.

  2. 02

    Agents check each stage

    Each stage agent returns findings with the evidence it relied on, so every warning points somewhere real.

  3. 03

    The orchestrator decides

    The orchestrator agent turns blockers and conditions into GO, CONDITIONAL GO or NO-GO. An AI review can add caution, never remove it.

  4. 04

    A person signs off

    Record the sign-off, then export the report as PDF, Markdown or JSON for the change board or the audit trail.

Why you can trust the call

AI can raise a flag.
It can never lower one.

The verdict comes from one small, readable rule. AI reviews can add conditions or blockers, but the decision is always rebuilt from those lists, so a confident model can never talk a NO-GO into a GO.

  • Findings cite the input they came from
  • Unknown evidence counts as missing, not as a pass
  • Every run traced with OpenTelemetry
How we handle security and your data
release_orchestrator.py
def decision_from(blockers, conditions):
    """Blockers beat conditions beat GO."""
    if blockers:
        return "no_go"
    if conditions:
        return "conditional_go"
    return "go"

Simplified from the real orchestrator in this repository.

Before and after

Release day, minus the guesswork.

RequirementsSpread across tickets and chatsParsed into criteria and risks
Test planningCoverage is a guessScenarios mapped to each requirement
API validationSomeone runs Postman, maybeGenerated tests run against staging
Code riskWhole-PR gut feelScored on changed lines, CI read in
Release callA meeting and a thumbs upA verdict with blockers, signed off

Questions

Straight answers.

What it does, what it does not, and where the line sits.

Is it really agent-based?

Yes. Each stage of a release has its own agent, and an orchestrator agent combines their findings. The stage agents run inside a harness with an allowed tool list and tracing, and can call Claude for deeper review. The final call follows one auditable rule, so AI can add caution but never wave a risky release through.

Does it deploy anything for me?

No. It recommends and records. Deployment stays with your pipeline and your people, and the CI repair agent only opens a draft pull request after a human approves it.

Where does the evidence come from?

From what you give it: requirements or Jira stories, an OpenAPI document, a pull request diff or link, and GitHub check results. Findings cite the input they came from.

What does a release-level verdict mean?

Group stories by hand or by Jira fix version. A missing, rejected or NO-GO story blocks the release, so one weak link is never averaged away.

How do you know the agents behave?

A known-case regression suite and DeepEval scorecard run in CI on every change, and tracing records each agent run so you can inspect what happened.

Put your next release through it.

One requirement, one OpenAPI document, one pull request. A few minutes later you have a verdict, the evidence behind it, and a report to share.

Talk to us

Bring us your next release.

Whether you want to pilot it on a real release, see it on your own stack, or explore working together, tell us a little about your team and we will get back to you.

  • Pilot on a real releaseRun your stories, API and pull requests through the agents and get a verdict you can check.
  • A walkthrough on your stackSee how it fits with GitHub, Jira and the CI you already use.
  • PartnershipsBuilding release or quality tooling? We are happy to compare notes.

Prefer plain email? kumar.mohit040289@gmail.com

What would you like to talk about?

Your message goes straight to our inbox, and your email app opens with a copy. We only use your details to reply.