We just open-sourced SBP (Stigmergic Blackboard Protocol), a protocol designed to let AI agents coordinate through environment-based signals rather than direct messaging or rigid orchestration.
The Problem: The "Orchestrator" Bottleneck
Most multi-agent systems rely on a central "brain" or complex message queues. This creates:
- Single points of failure: If the orchestrator hangs, the system dies.
- Brittle coupling: Adding a new agent often requires updating the logic of every other agent.
- Noise: Direct messaging scales poorly as the number of agents increases.
The Approach: Stigmergy
SBP is inspired by stigmergy: the mechanism ants use to coordinate. Instead of Agent A telling Agent B what to do, agents leave "signals" on a shared blackboard.
Key Mechanics:
- Signal Intensity: Agents post signals with varying strengths.
- Decay Curves: Signals aren't permanent. They fade over time (or through interaction), preventing the "blackboard" from becoming cluttered with stale data.
- Decoupled Awareness: Agents simply "sense" the blackboard and react based on their own internal logic.
SBP vs. MCP (Model Context Protocol)
We see these as complementary standards, not competitors:
- MCP defines Capabilities: "What tools can I use and what data can I see?"
- SBP defines Awareness: "What is the current state of the collective task and where is my help needed?"
🛠️ What’s Included
- Full Protocol Specification: RFC 2119 compliant.
-
Reference Server: TypeScript implementation (
@advicenxt/sbp-server). - SDKs: Ready-to-use libraries for TypeScript and Python.
- Pluggable Storage: Ships with in-memory support; extensible to Redis or SQLite.
- DevOps Ready: OpenAPI 3.1 spec and Docker support.
🔗 Links & Installation
- GitHub: AdviceNXT/sbp
-
npm:
npm install @advicenxt/sbp-server -
PyPI:
pip install sbp-client
Why we built this
We’re currently using SBP in production to manage non-linear workflows where agents need to "pick up where others left off" without a human or a script explicitly handing off the baton.
I'm happy to answer any questions about the protocol design, the math behind the decay mechanics, or how to implement a custom sensing loop!
Top comments (0)