Synapse

Configuration Language for Memory Systems

Synapse is a domain-specific language and runtime for building memory systems for AI agents. You write .mnm files that declaratively define schemas, event handlers, queries, update policies, and channel connectors. Synapse compiles them into a multi-backend runtime backed by SQLite, Qdrant, Neo4j, and 40+ messaging platform connectors.

Why Synapse?

Building memory for AI agents means juggling relational stores, vector databases, knowledge graphs, LLM extraction pipelines, and data ingestion from dozens of sources. Synapse unifies all of this behind a single, readable configuration language.


config {
  storage: sqlite("./data/agent.db")
  vector: auto("qdrant")
  graph: auto("neo4j")
  embedding: openai("text-embedding-3-small")
  extractor: openai("gpt-4o-mini")
}
 
memory Fact {
  content: string
  subject: string
  predicate: string
  object: string
  confidence: float[0,1]
  created_at: timestamp
 
  @index subject
  @index predicate
}
 
on ingest(content: string) {
  content |> extract() |> store()
}
 
query Search(input: string): Fact[] {
  from Fact
  where semantic_match(input, threshold: 0.6)
    and graph_match(input, hops: 2)
  order by confidence desc
  limit 10
}

One file. Multi-backend storage. LLM-powered extraction. Semantic and graph search. Update policies. Channel ingestion from Slack, Discord, Telegram, and 37 more platforms.

Key Features

Define memory schemas, handlers, queries, and update rules in .mnm files. No boilerplate.

Declarative DSL

SQLite for relations, Qdrant/ChromaDB/Pinecone for vectors, Neo4j/Memgraph/ArangoDB/SurrealDB for graphs. Mix and match with named backends.

Multi-Backend

Built-in extract() and summarize() powered by OpenAI. Extern functions let you define arbitrary LLM-simulated tools.

LLM Integration

Ingest data from Slack, Discord, Telegram, Reddit, Email, Webhooks, and dozens more. Each channel triggers ingest and update pipelines.

40+ Channel Connectors

Confidence decay, conflict resolution, periodic maintenance. Define how memory evolves over time.

Temporal Updates

Combine semantic search, graph traversal, regex filters, and SQL in a single query. Score aliases let you build custom ranking formulas.

Hybrid Queries

Quick Start


# Build the CLI
cargo build --release -p synapse-cli
 
# Start a memory system
./target/release/synapse apply examples/hello.mnm
 
# Store a note
./target/release/synapse emit save '{"content": "Remember to buy milk"}'
 
# Query all notes
./target/release/synapse query GetAll

Ready to dive in? Start with the Getting Started guide.