Eight arms. One brain.
Open-source swarm coordination layer for autonomous AI agent fleets. Shared memory, model routing, budget-aware dispatch, and priority signaling — all through standard MCP tools.
Eight arms. One brain.
Open-source swarm coordination layer for autonomous AI agent fleets. Shared memory, model routing, budget-aware dispatch, and priority signaling — all through standard MCP tools.
go install github.com/chitinhq/octi/cmd/octi-pulpo@latest
Everything your agents need to work as a fleet.
Route tasks to optimal models by cost, capability, and governance tier.
Vector DB for accumulated knowledge. Redis for hot state. Every agent learns from every run.
Who's working on what. Assignment dedup, handoffs, claims with auto-expiring TTLs.
Report up (agent to EM to director), get direction down. Structured decision chains.
Agent A needs agent B's output. Wait, notify, unblock — automatic.
Collective knowledge across runs. Denial patterns, workarounds, solutions — all shared.
Heartbeats, blocks, completions. Real-time swarm awareness across every agent.
Push a directive, every agent sees it immediately. Sprint goals, freeze windows, hot fixes.
Agent Swarm
Claude Code · Codex · Copilot · Gemini
│ MCP (stdio JSON-RPC)
Octi Pulpo
Coordination · Memory · Routing · Signals
│
Redis (hot state) + Vector DB (cold knowledge)
│
Chitin Kernel (optional)
Policy · Telemetry · Invariants
Three layers. One platform.