You plan the work.
Your agents take the night shift.
Write the tickets before you log off. Claude Code, Codex or Grok picks them up on your own machines, and By Dawn checks every step. By morning, a pull request is waiting.
Your agents can write the code. But someone still has to
By Dawn does the babysitting and the checking. Picking the work is still yours.
Plan it like a team.
Tickets, epics and cycles live on one board. Write each ticket the way you’d brief a colleague. Say what’s broken and how you’ll know it’s fixed.
Hand it to an agent.
Drag it into Building and close the laptop. Claude Code, Codex or Grok picks it up on one of your machines and works through it one step at a time. If it needs a decision from you, it stops and asks.
Merge it at breakfast.
The pull request is there when you sit down. It comes with the plan it followed, the test results, screenshots and the review comments it has already dealt with.
Bring your own coding agent
Bring
See what each machine is working on.
A box under the desk, a Mac mini or a cloud VM. Pair it once and it takes whatever tickets you send its way.
A board your agents can read too.
Planning → Lark, the board agent →
Your team and your agents work from the same tickets, epics and cycles. Here’s one night on the Lighthouse project, from the first ticket to the last pull request landing in Review.
Each ticket runs as a mission.
A mission is the list of steps a ticket has to get through: plan, build, verify, ship. An agent can’t mark its own homework, so By Dawn runs each check itself. If a step won’t pass, or the agent needs you to make a call, it stops and asks.
- Plan
- ✓Plan agreed
- Build
- ✓Find the cause
- ✓Fix it, with a regression test
- Verify
- ✓Tests and build pass
- ✓New tests are stable
- ✓Review panel
- ✓Screenshots, Chrome and Safari
- Ship
- ✓Open the PR, CI green
- ○You approveMerge
Tests that hold up
The suites have to pass on the exact commit it’s proving. New or changed tests run 10 times before they count, so a lucky pass doesn’t slip through.
Attempts per step
Three tries at a step, then it stops and asks you, with the log attached. Nothing merges in the meantime.
Screenshots
It clicks through the app in Chrome and Safari and shows you what it saw.
Changed lines covered
How many of the changed lines your tests run, read from your CI’s coverage report and held to your team’s target.
What your agents read before they start.
Your team writes the docs, and the rules that really matter get saved as short memory notes. Each run reads both before it starts, with nothing to set up. When an agent learns something new, it suggests an edit and a person approves it.
Docs your agents read before they touch the code.
- Read at the start of each runA run starts with the pages you’ve marked essential and the ones that fit its ticket, plus a snapshot of the space it can search. It notes which revision of each page it read.
- Write them togetherThe editor has what engineers reach for: slash commands, callouts, Mermaid diagrams, tables and ticket links. Comments, page history and templates are there too, and you can turn on live co-editing for any space.
- Agents suggest, people decideWhen a run learns something, it comes back as a change-set on the ticket, with a diff and a check that it still applies cleanly. Approve it in Review or straight from the chat.
- Memory with a triggerA memory note is a short rule, concept or decision that says when it applies. A ticket only gets the notes that fit it, and you can see how many tickets each one reached.
- Kept up to dateMark a page verified for anything from a week to six months, and its owner gets a nudge when that runs out. A health check flags duplicates, orphans and broken links.
- Easy to take with youIt’s all Markdown. Export one page or a whole space, mirror a space to GitHub as a pull request, or bring your pages over from Confluence.
The planning side, up close.
One place for incoming requests.
- Email, Slack, Discord and Microsoft Teams
- A public request form and in-app feedback
- Problems your agents run into along the way
Plan further ahead than tonight.
- Epics and milestones, with progress you can see
- Cycles, and a planner to fill them
- Releases with readiness checks and a changelog
Lark drafts the plan. You apply it.
- Ask it what’s stuck or what shipped
- In Act mode, it drafts changes to the board
- Nothing changes until you apply them
Your tests, out where the team can see them.
Test cases, suites, runs and plans sit next to the tickets, epics and releases they cover. The suites run on your own CI and machines, not on our servers.
- HealthThe flaky and broken tests costing you the most, ranked by how many pull requests and how much CI time they’ve hit. One click puts a test in quarantine.
- Cases, step by stepWrite out the steps and what should happen at each one, for manual or automated cases. Press “Automate this case” and an agent gets a ticket to write the test.
- Runs, by a person or the AI testerGo through a build case by case yourself, or let the AI tester drive the browser and take a screenshot at each step.
- Tests your agents writeNew tests from your agents stay in quarantine until a person approves them. Build library reads your repos and proposes a starting library.
You stay in charge.
Approve the plan
Before any code gets written, a checkpoint can wait for you to agree the plan or pick a mockup.
Steer it mid-run
Send a note while it’s working. The agent takes it on board and keeps going.
Needs you
Questions from all your agents wait in one list, so nothing sits blocked without anyone knowing.
Watch it, or take over
Follow its feed, terminal, browser and changes as it goes. Pause it, stop it, or take over the browser yourself.
From pull request to release.
Pull requests and releases → Insights and routines → Integrations →
Grouped by what they need from you
Read the diff, the review threads and each commit’s test results without leaving By Dawn.
Readiness checks with teeth
It checks for open work, failing CI and migrations that haven’t landed. An agent drafts the release notes for you to accept, and then you publish to GitHub.
Where the work waits
Cycle time, work that’s been open too long, where agents wait on people, and estimated agent usage by model and project. There’s no leaderboard.
Agent work on a schedule
Give an agent a playbook and a schedule. The weekly dependency bump and the morning triage are good ones to start with.
Customer requests, tied to tickets
Link each customer request to the ticket that answers it, and see whether it’s open, shipped or cancelled.
The tools you already use
GitHub, Azure DevOps, Slack, Discord and Microsoft Teams. Bring your backlog from Jira, Linear, GitHub Issues or Asana.
Things you’ll want to know first.
Where does my code run?
On your machines. You pair a desk box, a Mac mini or a cloud VM, and a small By Dawn daemon runs on it. The daemon connects out to us and picks up the work you send it. Nothing on our side ever opens a connection into your machine, and your checkout of the repo stays there.
What does By Dawn hold?
The board, your docs and memory, the test library, and each mission’s log and proofs. Your Claude Code and Codex sign-ins stay on your machine and we never collect them. If an agent uses an API key instead, as Grok can, you can keep that key in your org’s encrypted secret store.
Which agents and models?
Claude Code, Codex and Grok today, with Cursor, Copilot and OpenCode on the way. You choose the agent, machine, model and effort for each ticket, and you can set defaults for a project or for the whole org.
What does an overnight run cost?
Your agents run on the subscriptions or keys you already have, and we never resell a token. Insights estimates agent usage by model, project and ticket, and the Agents page shows how close each machine is to your Claude plan’s limits.
What happens when an agent can’t finish?
Each step gets up to three attempts. If it still won’t pass, or the agent needs a decision, it stops and asks in Needs you, with its log attached. Nothing merges.
Who merges?
You do, unless you change it. A project can let By Dawn’s reviewer approve and merge instead. The agent that wrote the code can never approve or merge its own pull request.
Can we keep Linear or Jira?
Yes. Sync with Linear or Jira both ways, pull from them only, or import once and move over. You can import from GitHub Issues and Asana too.
How does it learn our build and tests?
It reads your repo to work out the test suites and the build and test commands, without running anything. If it gets them wrong, you can set them yourself for each repo.
Can we self-host By Dawn?
Not today. The agents run on your machines, but the board runs on ours.
All of By Dawn, page by page.
Plan
- Planning →The board your team and its agents share, from triage to release.
- Lark →The board agent. Ask it about the work, or have it draft changes for you to apply.
- Docs →Pages your agents read on every run. Any edit they suggest waits for a person.
- Memory →Short notes that say when they apply, sent only to the tickets they fit.
Build
Prove
Ship
- Pull requests and releases →Pull requests grouped by what they need from you, and releases you can’t force past a red gate.
- Insights and routines →Flow, cycle time, where agents wait on people, estimated usage, and agent work on a schedule.
- Integrations →GitHub, Azure DevOps, Slack, Discord, Microsoft Teams, and your trackers.
- Security →Your code stays on your machines; nothing on our side connects in.