---
title: "CoderHelm: AI coding agent for GitHub and Jira. Ship code, not tickets."
description: "Assign a GitHub issue or Jira ticket and get a pull request. Then CoderHelm keeps the code healthy: it fixes its CI, answers review comments, reviews PRs, turns alerts into fixes and answers questions about your code."
url: https://coderhelm.com/
updated: 2026-11-05
---

# Ship code, not tickets.

> Assign a GitHub issue or Jira ticket and get a pull request. CoderHelm reads your codebase, plans the change, writes it and reviews its own work, then keeps the code healthy after it ships.

CoderHelm is an AI coding agent for GitHub and Jira. Assign it an issue or a ticket and it opens a pull request in your repository, then it stays on the code: it answers review comments, fixes its own failing CI, reviews your team's pull requests and turns production alerts into fixes that a person approves. Teams can also ask it about their code and documents in Chat, and run it hosted or in their own AWS account.

## One ticket, followed all the way to production.

### Turn tickets into pull requests

1. **Assign the ticket.** Assign an issue or a ticket, or add a label. On GitHub, assign the issue to CoderHelm, add the `coderhelm` label or comment `/coderhelm`. On Jira, assign the ticket or add the label. It reads the ticket, your AGENTS.md and the code, then writes a plan with the files it will touch. [Tickets to pull requests](https://coderhelm.com/pull-requests/)

   *Product view: A Jira ticket assigned to CoderHelm, and the plan it wrote before touching any code.*

2. **Open the pull request.** It builds, tests and reviews its own work first. It writes the change on a `coderhelm/` branch, runs your build, lint and tests in a sandbox that holds no credentials, reviews its own work for up to three rounds, runs a security check and opens the pull request. [What happens during a run](https://coderhelm.com/pull-requests/)

   *Product view: A run in the CoderHelm dashboard: each stage from triage to the opened pull request, with the files it changed.*

### Keep pull requests healthy after they open

3. **Fix failing CI.** A check fails. It reads the logs and fixes the cause. It pushes again until the checks pass, stops when the same failure repeats and caps automatic pushes at eight. When an end to end test fails for a reason the change did not cause, it runs it once more and says so. [How it fixes CI](https://coderhelm.com/maintenance/)

   *Product view: A failing check on a CoderHelm pull request: the log line it read, the fix it pushed and the checks going green.*

4. **Answer review comments.** Ask for a change on its pull request and it makes it. It replies in the thread, pushes the fix and says what changed. It also resolves conflicts with the base branch, so the pull request does not go stale while it waits. [Maintenance before and after the merge](https://coderhelm.com/maintenance/)

   *Product view: A review comment on a CoderHelm pull request, its reply and the commit that made the change.*

5. **Review every pull request.** Add a label to any pull request and it reviews it. Not only its own. It reads the callers and tests around the change, comments on exact lines with suggestions you can apply, rates the risk and says why, and reviews again on every push. It learns from which comments your team acts on or dismisses, and shows what it changed and why. [Everything the reviewer does](https://coderhelm.com/reviewer/)

   *Product view: A CoderHelm review: a risk level with the reason, a comment on the exact line with a suggested fix, and earlier comments it resolved after they were fixed.*

6. **Merge and check the deploy.** Nothing it wrote merges without a person's approval. Auto merge is off by default. Turned on, it waits for a person's approval and green checks on the exact reviewed commit, then watches the deploy checks and reports only failures that are new since the merge. [After the merge](https://coderhelm.com/maintenance/)

   *Product view: An approved pull request merged on the reviewed commit, followed by the deploy check CoderHelm ran after it.*

### Catch problems in production

7. **Turn alerts into fixes.** A production alarm becomes a fix in the right repository. Route CloudWatch alarms to CoderHelm through SNS and a rule picks the repository. It analyzes the alert, opens a pull request for a person to approve, and pairs the recovery with the original alert. It also reads the log groups you choose every six hours through a read only role, and can open a fix for each new finding on its own. [Alerts to pull requests](https://coderhelm.com/alerts/)

   *Product view: A CloudWatch alarm routed to the right repository, the pull request CoderHelm opened, and the recovery paired with the alert.*

8. **Keep tests from blocking.** Flaky tests stop failing builds. Broken ones are never hidden. It reads your Playwright, JUnit and Cucumber reports and tells flaky tests from broken ones. With quarantine on, flaky tests stop failing builds, and its fix pull requests must pass twenty runs in a row. [Flaky test detection](https://coderhelm.com/tests/)

   *Product view: Tests sorted into flaky, broken, quarantined and passing from their last ten runs, and a fix pull request with twenty green runs.*

## Ask about your code. Get answers you can check.

Chat reads your team's repositories itself: it looks a repository over, searches the code, reads the files and follows the code graph. Every claim has a numbered citation with the real file path and lines, and it never asks you to paste what it can open.

- It reads everything you can see; type @ to narrow a question to one repository or knowledge collection
- Knowledge from Notion, Confluence, Google Drive, Sanity, GitHub, websites and files, kept in sync
- When nothing in your sources supports an answer, it says so

- [How Chat works](https://coderhelm.com/chat/)

*Screenshot: Chat answering a question about a repository: what it read, then an answer with numbered citations. Example data.*

## What it uses to get the change right.

- **[Checks your real systems before it guesses](https://coderhelm.com/data-sources/):** Read only lookups in your GraphQL API, CloudWatch Logs, PostgreSQL and Redshift, enforced in code and recorded with who asked and for which pull request.
- **[Plans you approve](https://coderhelm.com/pull-requests/#plans):** Describe the work in the dashboard. CoderHelm streams a plan, and each task you approve becomes an issue and a run, in order.
- **[Works your way](https://coderhelm.com/skills/):** Reads your AGENTS.md, the skills your team keeps in a repository and the MCP servers you connect, and keeps a memory you can see and edit.
- **[Guarded in code, not in a prompt](https://coderhelm.com/security/):** Allow, ask or deny for everything an agent does, with approvals. Its own branch, a build sandbox with no credentials, read only data access, secrets encrypted with AWS KMS and custom roles.

Works with GitHub, Jira, Microsoft Teams, AWS, Playwright, JUnit and Cucumber reports, 17 MCP servers, and Claude models through your own Anthropic API key or Amazon Bedrock. Build on it with the [public API](https://docs.coderhelm.com/api/) and [signed webhooks](https://docs.coderhelm.com/api/webhooks/) that tell your systems what it did.

## Gets better with your team.

CoderHelm learns from what happens to its work on your repositories, and everything it learns is written down where you can read it.

- **Learns from what happens next:** A pull request merged as written or after edits, closed or reverted. CI passing on the first try or needing fixes. Review comments acted on or dismissed, and a thumbs up or down on a review. Each outcome scores the work.
- **Reads past work before it writes:** Before a change or a review, it reads the closest past examples from that repository: the ones that went best, to follow, and the ones that went worst, to avoid.
- **[The reviewer learns from pushback](https://coderhelm.com/reviewer/):** When someone dismisses a review, argues a comment or merges anyway, it checks whether the review was wrong. If it was, it writes one lesson for later reviews and shows what it changed and why.
- **[Everything it learns, you can see](https://coderhelm.com/skills/#memory):** Lessons, memories, examples and each change to how it works are listed with the evidence behind them. Admins can delete a lesson, edit or forget a memory, and turn off any change.

- [How the reviewer learns](https://docs.coderhelm.com/getting-started/reviewer/#it-learns-from-pushback)
- [Memories in the docs](https://docs.coderhelm.com/concepts/memories/)

## Run it in your own AWS account.

Your code, tickets, logs and model calls stay in your account. The install guide sets it up from your browser and checks every step against your own install.

- One click opens AWS with every value filled in, and the guide follows the install as it runs
- Your GitHub App in one click, and a check that Claude answers on Amazon Bedrock
- Your own domain, set up in Route 53 for you or with DNS records to copy, checked live
- The same install checks in the guide, the `doctor` command and the dashboard's Setup page
- Updates applied from the app in one click, or automatically in your maintenance window, from the Stable or nightly Beta channel
- Quiet email: only when you need to act or something went wrong, plus an optional weekly what's new
- Every install and update checks the release's signature before anything runs
- Fully private, or hybrid: use CoderHelm's GitHub App and event relay, and see every event in and out on the Deliveries page
- One organization per install: nobody signs up on their own, and admins can require two factor sign in

- [Open the install guide](https://docs.coderhelm.com/self-host/install/)
- [More about self-hosting](https://coderhelm.com/self-host/)

The guide talks to AWS, GitHub and your own install, and counts anonymous page views with Google Analytics. Nothing you type and nothing about your install is sent to CoderHelm or Google.

Moving from CoderHelm cloud? [Talk to us](mailto:support@coderhelm.com?subject=Moving%20to%20self-hosted%20CoderHelm) and we'll help move your settings and data.

*Screenshot: The Setup page of a self hosted install: each check passing, or saying what is wrong and how to fix it. Example data.*

## Questions teams ask first.

### What is CoderHelm?

CoderHelm is an AI coding agent for GitHub and Jira. Assign it an issue or a ticket and it opens a pull request in your repository, then it stays on the code: it answers review comments, fixes its own failing CI, reviews your team's pull requests and turns production alerts into fixes that a person approves. Teams can also ask it about their code and documents in Chat, and run it hosted or in their own AWS account.

### How do I start a run?

On GitHub, assign the issue to CoderHelm, add the `coderhelm` label, or comment `/coderhelm` or `@coderhelm`. On Jira, assign the ticket to CoderHelm or add the `coderhelm` label (your admin can change the label).

### Does CoderHelm push to my main branch?

No. It writes every change on its own `coderhelm/` branch and opens a pull request. Merging is your call; the optional auto merge for reviewed pull requests waits for a person's approval and green checks.

### What happens after the pull request is merged?

CoderHelm stays on the code. It answers questions and review comments, reviews new pull requests that get the review label, turns routed alerts into fixes, keeps flaky tests from blocking builds and, with auto merge on, checks the deploy after the merge and reports only new failures.

### Which languages does it support?

It works on the repository as text and runs your own build and tests, so it is not tied to one language or framework.

### Which AI models does it use?

Claude models from Anthropic, through your team's own Anthropic API key. A self-hosted install can use Amazon Bedrock in its own AWS account instead. Admins can choose the models and set monthly token limits.

### Can we run it ourselves?

Yes. CoderHelm runs entirely in your own AWS account, installed from your browser with the [install guide](https://docs.coderhelm.com/self-host/install/). It is licensed under FSL-1.1-ALv2: you can run it, change it and self-host it for your own company, and each release becomes Apache 2.0 two years after it ships. See [self-hosting](https://coderhelm.com/self-host/).

### How much does it cost?

CoderHelm is in open beta. Each team brings its own Anthropic API key, so model usage is billed to your Anthropic account, and admins can cap it with a monthly token limit.

### Can it change data in my database?

No. Lookups are read only, enforced in code at several independent points: the query is checked, the transaction is read only and rolled back, and the database user cannot write.

---

Source: https://coderhelm.com/ · Markdown: https://coderhelm.com/index.md · Updated 2026-11-05 · Get started: https://app.coderhelm.com