Loom
Memory and coordination for AI agents
AI agents lose their context when a session ends. When you have multiple agents, they can't see what the others are doing. Loom gives them shared memory, smart task routing, and the ability to pick up where they left off.
Install in 2 minutes
One-click for your AI tool of choice
Add to claude_desktop_config.json
Settings → MCP → View raw config
codex mcp add loom
Works with any MCP client
How it works
Start a task
Tell Loom what you're working on. It tracks the context.
Pick the right agent
Loom suggests which agent fits best—by capability, cost, or speed
Save your progress
Checkpoint anytime. If something crashes, pick up where you left off.
# Start working (create + claim in one call)
loom work "Fix authentication bug" --agent claude-code
# Save progress (crash recovery point)
loom checkpoint "JWT validation implemented"
# Get routing recommendation
loom route lm-abc --strategy cheapest
→ Route to: gemini (score: 0.85, cost: $0.001/1k)
# Complete with evidence
loom complete lm-abc --evidence "commit abc123"How Loom compares
Different tools solve different problems
| Feature | Beads | Gas Town | Loom |
|---|---|---|---|
| Multi-agent coordination | No | No | Yes |
| Smart routing | No | Basic | Yes |
| Session memory | No | Yes | Yes |
| Crash recovery | No | Yes | Yes |
| Git sync | Yes | No | Yes |
| Ground integration | No | No | Yes |
| Cost optimization | No | No | Yes |
What you can do
30+ tools for coordinating AI work
Manage tasks
loom_work— Start working on something right nowloom_create— Create a task for laterloom_spawn— Break a task into smaller piecesloom_ready— See what's available to work onloom_complete— Mark something done
Route work
loom_route— Get a suggestion for which agent to useloom_agents— See all your configured agentsloom_analytics— See what's worked well in the pastloom_record_execution— Help Loom learn from results
Remember context
loom_checkpoint— Save your progressloom_resume— Pick up where you left offloom_get_resume_brief— Get a summary of what happenedloom_update_context— Note what files you changed, decisions you made
Plan ahead
loom_formulas— Use workflow templatesloom_discuss— Talk through preferences before startingloom_verify_plan— Check if a plan makes sense
Use the right agent for the job
Not every task needs your most expensive model. Use Claude for architecture decisions, Gemini Flash for mechanical refactoring. Loom tracks what each agent is good at and helps you pick the right one.