DevIntern is a self-hosted worker process designed to automatically turn ready tickets in a tracker into pull requests. It runs on your own machine, uses the coding agent and model keys you already have, and lets you move from supervising interactive agent sessions to reviewing diffs.
Purpose
The core purpose of DevIntern is to delegate coding tasks asynchronously. It picks up tickets marked as ready from your tracker (such as Jira, Linear, Trello, Asana, Azure DevOps, GitHub Issues, or plain markdown files), automatically routes them to the correct repository, and runs an implementation loop that ultimately opens a pull request. You only need to review the pull request diff, not the agent session.
Main capabilities
- Ticket pickup and routing: A daemon watches the backlog, routes each ticket to the correct repository, and runs the implementation loop locally.
- Coding agent integration: Supports Claude Code, Codex, Cursor, or OpenCode. Agent configuration takes only one line, and switching agents is extremely low-cost.
- Model and key autonomy: Use any model supported by your agent, billed through your existing provider contract, with no bundled models or token markups.
- Feasibility gating: Checks task clarity before implementation; vague specs are flagged back to the tracker with questions, avoiding confident but incorrect pull requests.
- Optional self-review loop: When automatic review is enabled, the agent critiques the diff itself, fixes the issues it finds, and checks again before handing it to a human.
- Review comments turned into commits: When unattended automation is enabled, reviewer feedback is picked up and handled on the same branch, with replies attached.
- Robustness: Persistent queues, crash recovery, provider rate limit detection, and pause/resume.
- Optional ingestion: DevIntern PM can turn prompts, logs, or Figma frames into optional ticket ingestion.
How to use
- Install:
bun install -g @getdevintern/code - Run the worker:
devintern worker - Interactive use: run in the terminal
devintern PROJ-123to process a single ticket. - Configure the agent: switch via environment variables such as
AGENT_HARNESS=claude-codeorAGENT_HARNESS=codex. - Connect a tracker: supports Jira, Linear, Trello, Asana, Azure DevOps, GitHub Issues, and markdown files.
- Optional: connect a relay to receive events within seconds without exposing a public endpoint.
Suitable scenarios
- Engineering teams want to automate repetitive coding tasks and reduce manual supervision.
- Individual developers or small teams want agents to handle tickets in the background without adding infrastructure.
- Organizations that need to keep their existing tracker, review process, and CI unchanged.
Limitations
- Unattended automation is a paid feature; interactive use is free forever.
- Team subscriptions start at $99 per month and cover ten developers; individual supporter licenses are a one-time purchase.
- A 14-day no-card Worker Pilot trial is available, with no task limits.
- Subscriptions have a 30-day money-back guarantee.
- Source code is available under the FSL license and converts to Apache-2.0 after two years.
- Credentials and data stay local; optional services receive only limited operational data, not source code, diffs, ticket bodies, review comments, or vendor credentials.
- Specific model support depends on your coding agent.