Skip to main content
Brings an agent’s session down from its VM to your machine and continues it in the agent’s backend CLI: Claude Code for Claude agents, Codex for Codex agents. murmur resume rebuilds the agent’s workspace under ~/.murmur/workspaces/, checks out every repo on the agent’s working branch, downloads the session, registers the Murmur MCP server, and launches claude or codex from the workspace root. The resumed session includes the full conversation and the same file layout. The backend CLI must be installed: npm install -g @anthropic-ai/claude-code for Claude, npm install -g @openai/codex for Codex. Use murmur ssh instead when you want to work on the VM itself rather than locally.

​
Synopsis

The first form resumes the agent locally. The second tears the local workspace down again for one slug, or for every resumed workspace when no slug is given.

​
Arguments

​
What resume does

Resuming an agent runs these steps:
  1. Reconstructs the workspace under ~/.murmur/workspaces/, keyed by the agent’s identity. Every repo in the agent’s workspace is checked out on the agent’s working branch, using a git worktree of a nearby local clone when one exists or a fresh clone otherwise, and then pulled.
  2. Downloads the agent’s session, refreshed from the live agent when it is reachable. The download runs in parallel with the repo checkouts.
  3. Registers the Murmur MCP server so murmur CLI calls and MCP tools work from inside the resumed session. For Claude, the entry is scoped to the workspace directory. For Codex, it replaces the single global murmur entry in ~/.codex/config.toml.
  4. Launches the backend CLI into the resumed session from the workspace root, the same working directory the agent had on its VM, so paths in the session keep resolving. Claude resumes the downloaded session file; Codex resumes the installed thread with codex resume. A short notice is injected as the session’s next message telling the agent its environment moved.
Every step before the launch is best-effort and per repo: a single repo that fails to materialize is reported on stderr and resume continues. If the session cannot be downloaded, resume starts a fresh session of the backend CLI in the reconstructed workspace instead.

​
Examples

​
Resume an agent locally

The backend CLI opens with the agent’s full conversation and its repos checked out on the agent’s branch.

​
Resume an agent in another account

​
Tear down one resumed workspace

Removes the local workspace for fix-auth. A checkout with uncommitted changes prompts before deletion.

​
Tear down every resumed workspace

​
Errors

Run murmur resume from inside a directory where murmur is configured (a checkout with a .murmur/murmur.yaml above it, or with MURMUR_CONFIG set). Resume points the resumed session’s murmur configuration back at that directory, so the CLI and MCP tools keep working. If no configuration is found, resume still rebuilds the workspace and launches the backend CLI, but murmur calls in the session will not work until you run murmur setup there.