I've been building Broodlink to solve the infrastructure gap in multi-agent systems. Most frameworks give you an LLM wrapper — I needed shared memory, task routing, audit trails, and the ability to actually know what my agents did.
It's 7 Rust services, 66 tools, ~28K lines. The recent v0.5.0 release adds knowledge graph memory — LLM entity extraction builds a relationship graph in pure Postgres (no Neo4j). Agents can traverse multi-hop relationships, and edges track temporal validity so outdated facts don't pollute results.
Other things that might be interesting technically: hybrid BM25 + vector search with temporal decay and reranking (~200ms), Dolt for git-like versioned memory, MCP + A2A protocol support, and the entire knowledge graph runs on the same Postgres instance — no new infrastructure.
Happy to answer questions about the architecture or why I chose Rust/Postgres/Dolt over the more common Python/Neo4j stack.
I've been building Broodlink to solve the infrastructure gap in multi-agent systems. Most frameworks give you an LLM wrapper — I needed shared memory, task routing, audit trails, and the ability to actually know what my agents did. It's 7 Rust services, 66 tools, ~28K lines. The recent v0.5.0 release adds knowledge graph memory — LLM entity extraction builds a relationship graph in pure Postgres (no Neo4j). Agents can traverse multi-hop relationships, and edges track temporal validity so outdated facts don't pollute results. Other things that might be interesting technically: hybrid BM25 + vector search with temporal decay and reranking (~200ms), Dolt for git-like versioned memory, MCP + A2A protocol support, and the entire knowledge graph runs on the same Postgres instance — no new infrastructure. Happy to answer questions about the architecture or why I chose Rust/Postgres/Dolt over the more common Python/Neo4j stack.