Engine

Your agents, on your machine. Running while you're away.

The engine is the free runner of ctxflows: it takes ready work from your boards and runs your own coding agents on your machine or server, each in an isolated working copy, following your workflows. ctxflows holds the context and the workflows; the engine does the work where your code lives.

Any agent, or several

The engine runs the coding agent you already use, Claude Code, Codex, Cursor, Gemini CLI or Grok, and several of them side by side.

Choose which agents a project uses on each machine and which one is the default, and mix them per task, each running headless. One set of rules for every agent instead of one per vendor.

Unattended runs from your boards

The engine picks up only ready work: nothing blocked, nothing waiting on an item it depends on, in the order your roadmap sets, and only as much at once as you allow. Your goals steer what your agents build next.

Items that reach a column with a workflow run in the background and stop where an approval waits for you. Set the order on the roadmap, make items ready on their boards, and see what depends on what in memory and search.

Isolated working copies, in parallel

Every run works in its own working copy of your repository, so several agents work at once without touching each other's changes.

Your own checkout is never touched, and a finished run leaves nothing behind on your machine.

Checks on your machine

The checks a workflow names run on your machine, in the run's working copy: your tests, your linter, your build.

A node that needs a check can't close until it passes. ctxflows learns whether each check passed; the output stays with you.

Delivered your way

When an item's work is done, the engine delivers it the way your project chooses: as a branch to review, or landed one change at a time with your checks.

A branch goes to your code host, ready for a pull request and your own review. Landed changes join your target branch in the order they finished, each checked first.

Sessions and your coordinator

On an always-on machine or server, the engine starts sessions of your coding agent for you, and you continue them from your phone or the web.

Start a session from ctxflows and pick it up in your agent's own app. One coordinator agent watches the engine and every run, so you talk to one agent instead of checking each session.

What stays on your machine

Your code, credentials, agent sessions and transcripts stay on your machine or server; ctxflows receives what the work produced.

Your code never enters ctxflows. What reaches ctxflows: the outputs each node produces, whether each check passed, and what each session used, so you see what each item cost.

Free, on your own AI plan

The engine is free in every plan, and your coding agents run on your own AI subscription, so ctxflows never bills for agent usage.

Install and run it with the ctxflows CLI, ctx: connect a repository to your project and your coding agents, start taking work, add another agent and see the engine's state.

Three items overnight, two approvals in the morning

Before you stop for the day, three items reach Ready on your Feature board, and the engine on your server takes it from there.

  1. The engine picks them up

    In the order your roadmap sets, each item gets its own working copy and the agent its task calls for.

  2. Each run loops until its checks pass

    The agents plan, build and run your tests on the server, repeating the build until every check passes.

  3. One run asks a question

    The third run needs to know which error message to show. It waits in Actions, and waiting costs nothing.

  4. Two approvals wait for you

    In the morning you open each one and see its outputs, its check results and its branch.

  5. You approve one and send one back

    The approved change lands with your checks; the other run starts again from your note.

  6. You answer the question

    The third run continues where it stopped, with your answer in its context.

By the time you have read your mail, one change has landed, one is back at work and one question is answered, and the journal shows every step.

Questions about the engine

Do I need the engine to use ctxflows?

No. Connect any coding agent over MCP and work from your project in your own session; the engine adds unattended runs on your machine or server.

Which coding agents does the engine run?

Connect any coding agent over MCP. Install the free engine to run Claude Code, Codex, Cursor, Gemini CLI or Grok unattended, or mix them per task.

Does my code leave my machine?

Your code never enters ctxflows. The engine works in copies of your repository on your machine and pushes branches only to your own code host. ctxflows receives outputs, check results and usage, never your files.

Can I run the engine on a server, as a self-hosted coding agent?

Yes. Run it on any machine or server you keep, with git and your coding agent installed. It connects out to ctxflows, so your agents keep working while your laptop is closed.

What does the engine cost?

Nothing. The engine is free in every plan, and your agents run on your own AI subscription.

Put your coding agents to work on your boards