Skip to content

Demo: Multi-Agent Handoff

Who: Developers switching between AI models (e.g., Claude → Gemini → Cursor) mid-session. Time: ~1 minute | Prerequisites: Node.js 18+, a MultiModel Dev OS workspace with some history


Starting State

You've been coding with Claude Code for an hour. You want to switch to Gemini for an audit pass, but you don't want to lose the session context.


Workflow

Step 1: Build the memory index

bash
npx multimodel-dev-os@latest memory build

What happens: Scans your codebase and writes a hash-compressed index to .ai/intelligence/memory.hash.json and a human-readable summary to .ai/intelligence/memory.summary.md. Both are gitignored.

Step 2: Compile the handoff spec

bash
npx multimodel-dev-os@latest handoff build

What happens: Compiles a token-compressed session context document at .ai/intelligence/handoff.md. This includes:

  • Project structure summary
  • Active tasks from TASKS.md
  • Recent memory state
  • Feedback log highlights
  • Proposal status

Step 3: View the handoff

bash
npx multimodel-dev-os@latest handoff show

Expected output: Prints the handoff document to stdout — ready to paste into your next agent session.

Step 4: Check workspace status

bash
npx multimodel-dev-os@latest status

What happens: Shows a compact dashboard of package details, framework signals, memory freshness, feedback counts, proposal states, and audit metrics.


What Gets Created

FilePurposeGitignored?
.ai/intelligence/memory.hash.jsonHash-compressed file index
.ai/intelligence/memory.summary.mdHuman-readable memory summary
.ai/intelligence/handoff.mdToken-compressed session context

Safety Notes

  • All commands are read-only except memory build and handoff build which write to gitignored .ai/intelligence/ paths
  • No project source files are modified
  • Handoff documents are designed to be ephemeral — rebuild before each agent switch
  • Secret-safety exclusions automatically skip .env, .npmrc, .keystore files

Cleanup

bash
rm -f .ai/intelligence/memory.hash.json
rm -f .ai/intelligence/memory.summary.md
rm -f .ai/intelligence/handoff.md

Next Steps

  • Paste handoff into your next agent — copy the output of handoff show into Claude, Gemini, or Cursor
  • Run a health workflow: npx multimodel-dev-os@latest workflow run repo-health
  • Refresh after more work: npx multimodel-dev-os@latest memory refresh

Released under the MIT License.