Self-hosted · your coding agent · your tracker
Leave tickets. Come back to pull requests.
DevIntern is a worker you run on your own machines. It picks up ready work from your tracker, runs the coding agent you already pay for, and opens a pull request. You review diffs, not agent sessions.
$ bun install -g @getdevintern/code
Try one task free. Pay when the worker keeps going without you.
How it works
How the worker runs
One daemon watches the backlog, routes each ticket to the right repository, and runs the implementation loop locally. Review comments return to the same worker instead of becoming another terminal session for you to resume.
Optional intake
Prompt, logs, Figma → DevIntern PM
- 1
01 · INPUT
Ready tickets
Tracker queue
3 queued - 2
02 · ROUTE
Worker
Routes across repos
- 3
03 · BUILD
Coding agent
Isolated worktrees
- 4
04 · CHECK
Quality gates
Auto-review optional
- 5
05 · HANDOFF
Pull requests
Ready for humans
2 open
Review loop
Comments return to the worker as commits on the same branch.
Need a ticket first? DevIntern PM turns a prompt, log, or Figma frame into optional intake for the worker.
The async dividend
Your attention does not parallelize.
Supervising interactive agent sessions is still your focus, all day: watching terminals, approving steps, holding context for every open run. DevIntern moves your involvement to checkpoints. Work happens in the background, and the only interruptions that reach you are the ones that deserve to: a spec that needs a human call, a PR ready for review, a comment already addressed.
Your morning, supervising sessions
- Five terminal tabs, each mid-session, each waiting on you
- Context held in your head for every open run
- Faster than typing the code, but still your attention, all day
Your morning, with DevIntern
- Overnight tickets arrived as pull requests, checked and described
- One spec flagged as ambiguous, waiting on a human call in the tracker
- Your day starts with diffs to review, not sessions to resume
No lock-in
Swap any piece at any time
Every competitor bundles the tracker, the agent, and the model into one lock-in; DevIntern deliberately bundles none of them. Model rankings reshuffle every few months, and each reshuffle makes the neutral layer more valuable.
Your tracker
Jira, Linear, Trello, Asana, Azure DevOps, GitHub Issues, or plain markdown files. Full lifecycle, not just reading titles: batch selection with JQL, status transitions on pickup and completion, and the implementation summary posted back as a comment with the PR link.
Your coding agent
Claude Code, Codex, Cursor, or OpenCode. The agent is one config line. When a better one ships next quarter, switching costs that line, not a workflow migration: the tracker wiring, the review loop, and the run history all survive the swap.
Your model, your keys
Whatever models your agent supports, billed on your existing provider contract. No bundled model, no token markup, no waiting for a vendor to adopt the new frontier model. Day-one support for every release, by construction.
AGENT_HARNESS=claude-code
AGENT_HARNESS=codex # that is the whole migration
Built for unattended work
How it holds up without you watching
Async delegation only works if the loop catches its own mistakes. Four mechanisms carry that weight.
Feasibility gate
Before implementing, the task is checked for clarity. Vague specs get flagged back to the tracker with questions instead of becoming a confidently wrong pull request.
Optional self-review loop
Turn on auto-review and the agent critiques its own diff, fixes what it finds, and checks again before a human sees it.
Review comments become commits
With unattended automation on, reviewer feedback is picked up and addressed on the same branch, with replies. The cycle does not end when the PR opens.
Survives real life
Persistent queue, crash recovery, and provider rate-limit detection with pause and resume. The unglamorous 20% that kills the DIY script is built in.
Interactive use is free forever
Run devintern PROJ-123 from your terminal whenever you want. Every tracker, every coding agent, the planning tool, optional auto-review. No time limit, no signup, no crippled features. One run proves the loop before you put the worker on your backlog.
- Runs on your machine: no infra, no admin approval, no rollout window
- All trackers and all agents included: Jira, Linear, Trello, Asana, Azure DevOps, GitHub Issues, markdown
- Your own coding agent and keys: nothing new to buy, nothing metered by us
- Source-available, so you can read exactly what it does before pointing it at your repo
Pay when it runs without you
The paid line sits exactly where the value does: the worker keeps going. It polls your tracker and its own pull requests with no open ports. Connect the optional relay for events within seconds, still without exposing a public endpoint. Team covers ten developers for $99 a month: less than a single engineer-hour.
- Ready-ticket pickup and review handling on your own server, devbox, or spare VM
- A team subscription from $99 a month: no per-repo metering, no seat spreadsheets
- Solo? A one-time supporter license covers unattended automation for one person's own repos
- 30-day money-back guarantee on subscriptions, no questions asked
- Early pricing: subscribers keep their price for as long as they stay subscribed
Read the source before you hand over a Jira token
DevIntern is source-available under the FSL: read it, audit it, self-host it. Your code, credentials, and diffs never touch our servers, and the source is how you verify that instead of trusting it. Each release converts to Apache-2.0 after two years, so the worst case is "stops improving", never "stops working".
Read the source on GitHub Where every credential lives, in full →
FAQ
The questions engineers actually ask
Why not just give my coding agent a tracker connection myself?
You can, and it works for one ticket while you watch. DevIntern is what you build after that demo: the queue, the branch and PR and status state machine, optional auto-review, review-comment handling, rate-limit recovery, and a record of what happened to every ticket. The source is available, so compare it to your script.
Is agent-written code good enough to merge?
Sometimes no, which is why there is a feasibility gate before implementation, an optional self-review loop for the diff, and an escalation path instead of a forced merge. Your reviewers stay the final gate, exactly as they are today.
You want my tracker token and push access?
They stay on your machines. Credentials live in a local env file and each one is sent only to its own vendor's API. Optional DevIntern services receive narrow operational data, never source code, diffs, ticket bodies, review comments, or vendor credentials. See the security page for the full inventory.
Another subscription?
Interactive use is free forever, with no time limit. You pay for unattended automation, and you can self-build from source if you would rather. The pricing page says so out loud.
What if you disappear?
The license (FSL) converts to Apache-2.0 two years after each release, the code is public, and everything runs on your infrastructure. The failure mode is "stops improving", not "stops working".
Do we have to change our workflow?
No. Same repo, same reviewers, same CI, same tracker. Pull requests arrive the way your team already reviews them, and one engineer can start alone without asking anyone.
Which trackers, agents, and models are supported?
Trackers: Jira, Linear, Trello, Asana, Azure DevOps, GitHub Issues, and plain markdown files. Coding agents: Claude Code, Codex, Cursor, and OpenCode. Models: whatever your agent supports, on your own keys.
Credentials and data flow, in full: the security page →
Put a worker on your backlog this afternoon.
Try one ticket from your terminal, then leave devintern worker running against the tracker and repositories you already use. Your reviewers stay in charge of what merges.
Free for interactive use. No signup, no time limit, no credit card.
Different seat? DevIntern for engineering managers, product managers, and founders.