multi-repo git dashboard

Know which repo needs attention.

Fleet Radar is FluxGit's local attention stack for developers juggling many repositories. It highlights dirty worktrees, branches ahead or behind, divergence, stale snapshots and recovery signals, runs guarded actions across many repositories at once, and shows which coding agents are working in each one.

FluxGit Fleet Radar with two coding agents working in two repositories, two agent proposals waiting on the orbit-web row, a warning that Claude Code is changing files in orbit-api with no proposal waiting, and the scheduled fetch setting switched off
Real FluxGit interface · synthetic demo data · agents per repository, proposals on their row, scheduled fetch off by default

The problem: multi-repo work hides risk.

Developers often have several repos, worktrees or related projects open across a feature. Without a dashboard, drift and dirty work surface too late: during push, release, merge or context switch.

Current scope

Fleet Radar is in beta. It works on the local repositories you save, open or recently used, and it refreshes when you ask. Scheduled fetch is opt-in and off by default: once enabled, it fetches saved repositories while FluxGit runs, up to 10 per cycle, one at a time, and only updates remote-tracking refs. It is not company-wide monitoring, telemetry or auto-discovery of every repository on disk.

How FluxGit helps.

signals

Attention stack

Repos are sorted by signals that matter: WIP, divergence, behind/ahead state, conflicts and recovery context.

refresh

Throttled refresh

Local scans never contact a remote. Refresh is explicit, and the optional scheduled fetch never pulls, pushes, changes HEAD or stages files.

recovery

Jump to safety

When a repo needs action, FluxGit points back to the relevant safety or recovery surface instead of hiding it in a list.

Act on many repositories without losing track of any.

Select repositories and FluxGit plans the batch first: a live preflight marks each one ready or blocked, you run it on the ready ones, and every batch is recorded.

batches

Fetch, pull and push

Plan a fetch, pull or push across the selection. Pull is fast-forward only and push never forces. If Git state changes after the plan, FluxGit asks for a new one.

branches

Branches across repositories

Create a missing branch at each repository's HEAD or switch branches across the fleet. Delete a merged branch across many repositories at once, with a restore point for every tip. Restore recreates the branch at its recorded tip; tracking settings are not restored.

digest

What changed across the fleet

See what changed across your whole fleet since your last review, newest first, from the reflogs of HEAD and local branches, with history rewrites flagged.

identity

Moved or replaced repositories

Each saved repository remembers its root commit. Fleet Radar notices when a saved repository was moved or replaced, blocks writes to it and helps you locate it again.

submodules

Drifted submodule pointers

Record a drifted submodule pointer in its parent with a commit that contains only that pointer, or return the submodule to the recorded commit with a restore point, at any depth.

Which coding agents are working where?

Each repository row shows the coding agents working in it, and how FluxGit knows: a running process, a session log or an editor workspace. Agents connected to FluxGit's MCP server are recorded by the server itself. When an agent is changing files with no proposal waiting, the row says so. Agent proposals waiting for you show up on the repository they belong to, and Review opens that proposal.

From your phone, continue a supported agent conversation or open an authorized desktop terminal. Explore mobile agent control

How it knows

Detection is local and checks every 5 seconds while the view is open. It reads presence, not intent: FluxGit never reads or returns prompts, and agent names are what each tool reports about itself. The same agents appear on the commit graph, on the row where each one is working.

Privacy and security posture.

Fleet Radar is local-first repository awareness. It does not send repository names, paths, remotes, branch names, commit IDs or file names to licensing or feedback by default. Its MCP tools are read-only and never fetch on an agent's behalf.

Related features.

  • Safety rails handle risky operations after Fleet Radar identifies attention.
  • Reflog recovery gives rescue paths when a repo shows recovery signals.
  • MCP agent Git lets agents inspect fleet state and recent fleet activity read-only.
  • AI coding agent control connects Fleet Radar to agent conversations, remote terminals, mobile access and reviewable Git actions.
  • Submodules and worktrees covers parallel agent attempts and submodule pointers.
  • FluxGit Mobile can show explicitly shared repository state from a paired desktop during the private preview.

Recorded from the application

Eight tracked repositories and seven asking for attention, ordered by what is worst: conflicts, then divergence, local work and drift. The attention stack reads last-observed local snapshots, so counts stay labelled stale until you refresh or fetch. Fetching is manual unless you turn on scheduled fetch. Synthetic demo data.