Fleet runs

A fleet run coordinates agent work across multiple git repositories with explicit barrier sync. Run repo A, wait until it finishes, then run repo B.

This is different from modular projects, which coordinate multiple folders inside one project manifest and one run_id.

At a glance

• Fleet manifest at ~/.pytxo/fleets/<id>.toml
• Nodes are repos with their own cmd, agents, and optional depends_on
• Waves run in topological order; nodes in the same wave can run in parallel
• Progress is recorded in the catalog; use pytxo fleet status (or MCP) to watch fleet runs today. A dedicated Desktop fleet panel is coming later.

Create a fleet

pytxo fleet init my-fleet \
  --add ~/dev/backend --cmd "echo backend" \
  --add ~/dev/frontend --cmd "echo frontend"

Preview the plan:

pytxo fleet dry-run --id my-fleet

Run

pytxo fleet run --id my-fleet

Continue later waves even if one node fails:

pytxo fleet run --id my-fleet --continue-on-error

Check history:

pytxo fleet status --id my-fleet

Pytxo Desktop

Fleet progress is available from the CLI and MCP today. Use pytxo fleet status until a Desktop fleet panel ships.

MCP from Cursor

The MCP tools pytxo_fleet_run and pytxo_fleet_status let your IDE start or inspect fleet DAGs without leaving the editor. See MCP from Cursor.

Example manifest

[fleet]
id = "release-train"
name = "Release train"

[[node]]
id = "backend"
repo = "/home/dev/backend"
cmd = "pytxo run --agents 2 --cmd 'npm test'"
agents = 1

[[node]]
id = "frontend"
repo = "/home/dev/frontend"
cmd = "pytxo run --agents 2 --cmd 'npm test'"
agents = 1
depends_on = ["backend"]

Back: Execution domains