From ticket to pull request. You just approve.
Flusend Agent picks up a card from your Linear, fixes it in an isolated worktree, runs your official test suite and opens the PR. Merging is the approval, and it triggers the deploy your repository already has.
- ✓Isolated worktree per ticket
- LinearGitHubSentry
- ✓Existing tests untouched
Isolated worktree
your branch is never touched
Agent + 12 skills
root cause, fix, new test
Official test suite
4,265 tests · green · 5 min
Protected paths
migrations, terraform, .github
PR #56 opened
diff, trail and cost in the UI
Waiting for approval
merge = deploy
From ticket to deploy, with you in charge
Six steps. The agent only touches code after you authorize the plan, and only reaches the environment after you approve the PR.
Open the card in Linear or from the chat, typing or speaking. The agent label is all it takes.
The agent reads the real repository code in an isolated worktree and understands the cause.
It shows the implementation or fix it will make. Nothing changes before you authorize.
Once authorized, it builds the feature or bugfix, writes the test and runs your official suite.
Opens the PR on your GitHub with diff, trail and cost. Reviews go back to it.
With the PR approved, the merge triggers your repository's deploy to the environment and the card is done.
Export to PDF fails without patient address
Built for PRs you can actually merge
An agent that tampers with tests to get green delivers nothing. These rules are set per repository and shown on every run.
@Testvoid totalsThreeItems() {- assertEquals(3, order.total());+ assertEquals(2, order.total());}
+@Test+void appliesDiscountOnce() {+ order.apply(coupon);+ assertEquals(90, order.total());+}
Existing tests untouched
The agent writes new tests and adds cases, but never edits or deletes what was already there. Tools refuse, and the diff check blocks delivery. You can allow it per card when it makes sense.
Protected paths
Migrations, infrastructure, pipelines and secrets stay out of reach. A change in a protected path blocks the PR before the push.
Doubt becomes a question
An ambiguous ticket is not a guess: the agent comments its questions on the card and resumes the same session when you answer.
Limits per agent
Turn and cost ceilings, blocked commands and allowed tools, all editable in the UI.
Audit trail
Every run records what was read, run and changed, what was blocked and how much it cost.
Isolated in Docker, if you want
Running directly on the machine, the blocks are a guardrail. In the container, the agent only sees the data volume and what you mount on purpose.
The work doesn't end when the PR opens
That's where most agents stop. Flusend Agent keeps going until the card is really closed.
Reviewer · line 42
The empty-list case isn't handled here.
Flusend Agent
Handled and covered by a test. Same PR updated (round 1 of 3).
GitHub
PR #56 merged. Card MED-342 moved to Done.
Sentry
48 h without an occurrence on the release with the fix.
- Merge rate
- 3 of 4
- Time to PR
- 23 min
- Rework after merge
- 0 of 3
small sample: we show the count, not a percentage
Reviews go back to the agent
Inline comments and reviews, from people or review bots, go back to the agent on their own, and it updates the same PR. With a round limit.
Sentry checks production
A production error becomes a card in one click, with the backlog checked first. After the merge, 48 h without an occurrence resolves the error in Sentry; an occurrence in a release that already has the fix reopens the card.
Honest metrics
Merge rate, rework after merge, time to PR and usage per merge, always with the sample size next to it.
More than fixing bugs
All in the same UI, with your approval before any card is created.
Card draft
Export to PDF fails without patient address
Reproduce with a patient without an address, handle the null field and cover it with a test.
Chat, with voice
Describe the problem out loud. The text lands in the box, you review it, and get the card draft: create and send, create only, edit or discard.
Requirements
Documents become a list of requirements with their source, checked against the backlog before becoming cards.
New feature
A conversation becomes a plan read from the real code. You approve, the card is born and goes to the agent.
Errors (Sentry)
The queue of open errors, with you deciding what becomes a card, plus the follow-up after the merge.
Automations
Recurring task, errors above a threshold or cards by label. Every automation starts paused, has a cap per trigger and can be dry-run.
Agent inside Linear
Delegate or mention Flusend Agent on a card and follow progress, questions and the PR link without leaving Linear.
Agents and skills
Agent profiles and engineering skills editable in the UI. 12 built-in skills, yours on top.
Board, runs and PRs
Your whole Linear on a board with Send to agent on any card, the trail of every run, and PRs with the diff to approve.
Cloud coding agent vs. Flusend Agent
Same idea, opposite decisions on everything that matters at merge time.
| Cloud coding agent | Flusend Agent | |
|---|---|---|
| Where it runs | On the vendor's cloud | Wherever you want, Docker included |
| Who runs the tests | The agent says it passed | The orchestrator re-runs the official suite |
| Existing tests | The agent can change them | Protected: it only adds |
| After the PR | Ends at the PR | Reviews come back; Sentry checks after merge |
| Work queue | Its own UI | Your Linear, with one label |
| Cost | Per-seat subscription | The model cost, visible per run |
Talk to a specialist
Tell us about your repository and your test suite. We set it up together and you approve the first PR.
What you get
- ✓Install wherever you want, Docker included, unlimited repositories
- ✓Linear, GitHub and Sentry connected
- ✓Editable agents and skills, 12 built-in skills
- ✓Existing tests protected and blocked paths per repository
- ✓Parallel runs
- ✓Direct support from the people who built it
How model cost works
Bring your own Claude
The agent uses your Claude Code subscription or an Anthropic API key. Nothing goes through a Flusend server.
Cost per run, on screen
Every run shows the API-price equivalent. A real fix in a repository with 4k tests came in near US$ 2.
Ceiling per agent
Cost and turn limits per agent profile. Past the ceiling, the run stops and you decide.
What people usually ask
Flusend Agent is a software engineering agent that turns a ticket into a pull request ready for approval. It reads the real repository code, shows the plan, implements the fix or feature in an isolated worktree, runs the official test suite and opens the PR on GitHub. The merge triggers the deploy the repository already has. It integrates with Linear, GitHub and Sentry and runs wherever you want, Docker included.
Node 22.13 or newer, pnpm, git, an authenticated gh, and Claude Code installed and logged in (or an Anthropic API key). Or Docker: the container gets its credentials from .env and the agent is confined to it.
With any repository that has a test command. The first repository in production is Java with Spring and Maven, with over 4k tests; the UI and server are TypeScript.
It goes to the model during the run (your Anthropic account) and to GitHub when the PR is pushed. There is no Flusend server in between: the database, clones and worktrees stay with you.
No. Without a Linear key, Flusend Agent uses a built-in queue. With Linear, a personal key with read, create issues and create comments is enough.
It works in its own worktree; your folder and your branch are not touched. Protected paths block the push, and a red suite never ships. Deploy only happens if you merge.
The agent comments its questions on the card and stops. When you answer, it resumes the same session where it left off.
Send the first card.
Tell us about your repository and your test suite. We set it up together and you approve the first PR.
Talk to a specialist