swafra 0.2.0

Leiden-chunked, graph-linked semantic memory as an MCP server for Claude, Copilot, and any MCP-compatible AI

swafra

Semantic memory for AI — ingest anything, retrieve what matters.

94.7% recall_all@10 on LongMemEval — the standard benchmark for long-term memory in AI assistants.

Works as an MCP server with Claude Desktop, Claude Code, VS Code Copilot, and any MCP-compatible AI.

---

Install

pip install swafra

Or with Node.js:

npm install -g swafra

---

Connect to Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "swafra": {
      "command": "swafra"
    }
  }
}

Restart Claude Desktop — the tools appear automatically.

---

Connect to Claude Code

claude mcp add swafra swafra

---

Connect to VS Code (Copilot)

Add to .vscode/mcp.json in your project:

{
  "servers": {
    "swafra": {
      "command": "swafra"
    }
  }
}

---

What you can do

Once connected, Claude can use swafra to remember and retrieve anything:

"Remember this meeting transcript: ..."
"What did we decide about the API design?"
"What are my editor preferences?"
"Forget everything from the project X sessions"

---

Tools

Tool	What it does
add_knowledge	Store text — chunked, embedded, and graph-linked
search_knowledge	Find relevant chunks by natural language query
get_context	Search + graph walk combined (recommended)
graph_walk	Explore connected chunks from a starting point
list_sources	See everything stored
delete_source	Remove a source and all its data

---

How it works

1. Chunking Text is split into semantically coherent chunks using Leiden community detection — a graph algorithm that groups sentences by topic. Falls back to conversation-aware chunking if Leiden deps aren't available (Python 3.13+).

2. Local embeddings Uses fastembed (ONNX, CPU-only, no API key) with BAAI/bge-small-en-v1.5. Falls back to deterministic hash vectors if fastembed isn't installed.

3. Hybrid retrieval 4-signal fused scoring: BM25 + vector cosine + entity/date overlap + character n-gram. Returns the best chunk per source so you get diverse, non-redundant context.

4. Knowledge graph Chunks are connected with sequential (next/prev), similarity, and entity co-occurrence edges. Graph walk expands retrieval beyond what search alone finds.

5. Storage JSON files in ~/.scimap/. No database, no server, no cloud.

---

Benchmark

94.7% recall_all@10 on LongMemEval-S — 500 questions across 6 categories, 53 sessions each.

Category	recall_all@10
knowledge-update	100.0%
single-session-user	100.0%
single-session-preference	100.0%
single-session-assistant	100.0%
temporal-reasoning	99.2%
multi-session	93.3%

Full benchmark details and reproduction steps →

---

Environment variables

Variable	Default	Description
SCIMAP_DATA_DIR	~/.scimap	Where knowledge is stored
SCIMAP_EMBED_MODEL	BAAI/bge-small-en-v1.5	Embedding model

---

License

MIT — github.com/kunal12203/swafra