VaultRun + Flowd

Two local-first tools for different layers of automation. Flowd learns your file workflows on your machine. VaultRun gives AI agents a secure place to run code. Use both when you want approval-driven local automations and audited sandbox execution.

how they fit

  flowd                         VaultRun
  observe filesystem              Docker sandbox per session
  detect rename/move patterns     run_command · MCP · audit trail
  suggest → you approve           agents · scripts · CI
  flowctl undo                    network off by default

typical agent flow

1. Agent calls flowd_list_suggestions via VaultRun MCP (opt-in)
2. Agent calls flowd_explain_suggestion — you inspect the why
3. You approve → flowd_approve_suggestion
4. For heavier work: create_session + run_command in VaultRun

enable in MCP

$ cargo install --path crates/flow-cli   # from github.com/nickvd7/flowd
$ flowctl setup --watch ~/Downloads
$ flow-daemon &

$ MCP_FLOWD_ENABLED=true \
  FLOWCTL_PATH=flowctl \
  VAULTRUN_BASE_URL=http://localhost:8080 \
  VAULTRUN_API_KEY=vr_your_key \
  ./vaultrun-mcp

what is not bridged (v1)

• Flowd run / dry-run — still local via flowctl; no HTTP API in Flowd
• VaultRun does not observe your desktop — that stays in flow-daemon
• Both run on the same host as vaultrun-mcp when using flowd_* tools