Launched this week
Cosmonapse
An agent-to-agent protocol built for harness engineering
13 followers
An agent-to-agent protocol built for harness engineering
13 followers
Cosmonapse is an event-driven agent-to-agent protocol modeled on a nervous system. Agents are pure functions no base class, no lifecycle. Orchestration is a role, not a bottleneck: any node can dispatch, any node can react. Harnesses are built from typed signals instead of loops tool calls (incl. MCP), shared memory, human approval, and retries each compose as a node or hook around a stock model call. Same code scales from in-memory tests to NATS/Kafka. MIT. Python + TypeScript SDKs + CLI.
Typed signals instead of loops is a clever way to wire up retries and MCP tool calls, and the same harness running on NATS without rewriting feels like the kind of thing I'd actually reach for in a side project.
@tuerrv4Β Thanks! The topology part is where it gets fun, as everything is a node reacting to signals, complex harness shapes stay cheap. Fan a task out to competing bidders, put an approval gate in front of just the destructive tools, chain a retry policy behind a capability router... each is one more node on the bus, never a rewrite of a loop. And it's fully LLM-agnostic, the model call is a stock function the protocol never touches, so each node can run whatever it wants. Cheap local model for routing, a big one for the actual thinking, all in the same harness.