Shared local state
Keep projects, blockers, screenshots, and run history in one local tracker so you can leave the loop and return without rebuilding context.
Local-first orchestration for coding agents
Maestro keeps your local queue, MCP handoff, and control center aligned so you can offload repetitive execution, let agents keep moving, and return to live context instead of rebuilding the story from scratch.
1
Install
npm install -g @olhapi/maestro 2
Bootstrap workflow
maestro workflow init .Live control center
Hand work off without losing local context
Maestro earns the handoff by keeping the tracker, MCP bridge, and runtime supervision on the same machine. The loop can keep moving without turning into a black box.
Live MCP handoff
Expose that same queue through `maestro mcp` so your agent sees the work and state the daemon is actually supervising.
Visible runtime
Let Maestro keep routing ready issues while you watch retries, sessions, logs, and live state from the control center.
See when to step in and when to leave the loop alone
The walkthrough is about developer re-entry: read queue pressure, route work, inspect a single run, and get back out without babysitting every issue.
See whether the loop is healthy before you interrupt it
Overview keeps running agents, retries, throughput, and board pressure visible so you can decide whether to step in or let the queue keep moving.
- Retry trends tell you whether the loop is drifting.
- State totals make backlog pressure obvious.
- Live runs and pending retries are one click away.
Adjust the queue without stopping the rest of the work
The board keeps blockers, live sessions, token burn, and the shared composer together so you can reroute work and get back out quickly.
- Drag issues between lanes or inspect them in place.
- Open the shared composer and dictate issue descriptions with browser speech input.
- Filter by project, state, and search without flattening the board.
- Spot blocked or live issues before they become blind spots.
Step back into one issue with the full trail intact
When one run needs attention, the issue page keeps transcript, blockers, screenshots, branch state, and commands in one dense control panel.
- Execution state, screenshots, blockers, and follow-up commands stay adjacent.
- The transcript shows command progress and agent messages together.
- Attach or remove local images without leaving the issue.
Watch parallel work without opening every issue
Sessions is the fast path when multiple repos or issues are moving and you only want to intervene where the loop is actually stuck.
- See session identifiers, runtime churn, and last activity.
- Track retries and investigate stalled runs quickly.
- Keep the control-plane view even when agents are mid-flight.
See whether the loop is healthy before you interrupt it
Overview keeps running agents, retries, throughput, and board pressure visible so you can decide whether to step in or let the queue keep moving.
- Retry trends tell you whether the loop is drifting.
- State totals make backlog pressure obvious.
- Live runs and pending retries are one click away.
Adjust the queue without stopping the rest of the work
The board keeps blockers, live sessions, token burn, and the shared composer together so you can reroute work and get back out quickly.
- Drag issues between lanes or inspect them in place.
- Open the shared composer and dictate issue descriptions with browser speech input.
- Filter by project, state, and search without flattening the board.
- Spot blocked or live issues before they become blind spots.
Step back into one issue with the full trail intact
When one run needs attention, the issue page keeps transcript, blockers, screenshots, branch state, and commands in one dense control panel.
- Execution state, screenshots, blockers, and follow-up commands stay adjacent.
- The transcript shows command progress and agent messages together.
- Attach or remove local images without leaving the issue.
Watch parallel work without opening every issue
Sessions is the fast path when multiple repos or issues are moving and you only want to intervene where the loop is actually stuck.
- See session identifiers, runtime churn, and last activity.
- Track retries and investigate stalled runs quickly.
- Keep the control-plane view even when agents are mid-flight.
See whether the loop is healthy before you interrupt it
Three commands to start the loop and let it keep moving
Install the runtime, write the repo contract once, then launch the daemon so Maestro can keep reading ready work while you move on to review, coding, or planning.
Install
Install the local runtime once, then use the same command in every repo you want to supervise.
npm install -g @olhapi/maestroBootstrap workflow
Setup repo contract once so Maestro can keep making the same handoff decisions while you are elsewhere.
maestro workflow init .Start orchestration
Start the daemon and embedded control center so the queue keeps moving while status, retries, and sessions stay visible.
maestro runDocs organized around the handoff and the return
Start with install and quickstart, then use the docs to supervise the loop, tune its behavior, or debug it when something drifts.
Install and quickstart
Get from install to a live local loop, then learn what keeps running after the initial handoff.
Workflow configuration
Tune how much autonomy Maestro gets, how it retries, and which guardrails stay in place.
Operations and observability
Check queue health, inspect live runs, and debug the daemon without guessing what happened.