Autonomous coding agents inside your Jira and GitHub.
Mark a ticket Agentic and a Claude agent reads the requirements, writes the code, runs the tests, opens a PR, and reviews itself. You keep control of the two things that matter: review and merge.
Well-specified tickets shouldn't need a human babysitter.
For a large share of the backlog, the human reads the ticket, drives an AI through the change, opens a PR, and shepherds review. That's process — not judgment.
AI tools live outside your workflow
Separate dashboards, separate habits, separate review. Adoption dies the moment your team has to context-switch to a tool that isn't Jira and GitHub.
Autonomy you can't trust
Either you babysit every step, or a bot merges code nobody reviewed. Neither is acceptable for a real codebase with real consequences.
No audit trail
When an agent ships a change, can you prove what it did, which model ran, what it cost, and who approved the merge? Usually not.
How it works
One Jira project. Two human gates that never move.
Agentic and human work converge at code review — same process, same merge, same QA tail.
- 01
Triage: mark a ticket Agentic
Human gateA human decides which tickets run autonomously — right inside your Jira board. No new tool.
- 02
The agent reads the requirements
It pulls the ticket context, asks a structured question if something's genuinely unclear, and plans the change.
- 03
Writes code and runs the tests
In an ephemeral, isolated container with your toolchain — git, gh, your test suite.
- 04
Opens a pull request
A normal PR on your repo, linked back to the ticket, with the full transcript and diff attached.
- 05
Self-reviews, up to 5 rounds
A reviewer agent finds issues and dispatches a fixer. It keeps tightening — then hands off regardless.
- 06
Human review and merge
Human gateYour team reviews the PR and merges. Agents propose; people approve. Merge is a human-only gate.
Features
Velocity without giving up control or auditability.
Lives in Jira + GitHub
No new tool, no separate board. Agents coordinate through your existing workflow states.
Tamper-evident audit trail
Every action, transcript, diff, and review round is hash-chained and verifiable per ticket.
Model & effort per ticket
Pick the Claude model and reasoning effort for each ticket, with exact cost tracked and attributed.
Fleet dashboard
Live view of every agent — role, status, current ticket, and today's spend.
Self-review loop
A reviewer agent finds issues and a fixer resolves them — up to five rounds before handoff.
Open source
Free to self-host and use internally. No bot ever merges your code.
Building in public
Watch it get built.
The roadmap, the changelog, and the numbers — all in the open.
Roadmap
View all →- M1
Ticket Agent MVP
In progress- Monorepo scaffold + quality toolchain
- Local stack (docker-compose)
- Jira webhook ingestion + state machine
- Developer agent (implement → test → PR)
- Per-ticket model + reasoning-effort fields
- Swagger API docs
- M2
Agent runtime
Planned- Ephemeral, isolated runs per ticket
- Evidence store (transcripts, diffs, findings)
- One image, three roles: developer, reviewer, fixer
- Agent harness backends beyond Claude — Codex, Gemini CLI, OpenRouter
- M3
Self-review loop
Planned- Reviewer agent finds issues
- Fixer agent resolves them
- Up to five rounds, then hand off
- M4
Dashboard UI
Planned- Fleet view — live agent status + spend
- Ticket timeline + run detail
- Transcript + diff viewers
- M5
Hardening + audit anchors
Planned- Tamper-evident hash-chain verifier
- External audit anchoring
- Production hardening
- M6
End-to-end testing in the agentic environment
Planned- E2E test suites run inside the agent runtime
- Agents validate changes against real flows before handoff
- Test results attached to the run and ticket
Latest updates
See all →Get early access to ticketagent.dev
We're building in public. Join the waitlist for early access and the occasional product update — no spam.