Perseus Vault: Persistent Agent Memory Engine

Perseus Vault (formerly "Mimir"/"Mneme") is a persistent agent memory engine with structured entities and FTS5 hybrid search. It achieves 72% accuracy on LongMemEval oracle split — beating FullContext by 26 percentage points — while using no embedding API.

Key Features

• Structured entity storage with tags, categories, decay, links, verification
• FTS5 hybrid search — semantic keyword matching without embeddings
• HTTP/SSE MCP transport — connects to a running perseus-vault serve instance
• Zero embedding cost — all retrieval is local SQLite FTS5
• Proactive recall hooks (recall_when) for agent-driven memory

LongMemEval Results (oracle split, 50 temporal reasoning questions)

LLM: gemma-4-26b-a4b-it via OpenRouter

Setup

# Start Perseus Vault
perseus-vault serve --db /tmp/perseus_vault_bench.db --transport sse --port 8766 &

# Run benchmark
uv run python scripts/run_full_benchmark.py \
  --benchmark longmemeval --split oracle \
  --systems perseus_vault --num-samples 50

Links

• Repo: https://github.com/Perseus-Computing-LLC/perseus-vault
• Install: download a release binary from https://github.com/Perseus-Computing-LLC/perseus-vault/releases

(Renamed from "Mimir": the memory engine this benchmark targets was renamed to Perseus Vault. The system module is now perseus_vault — invoke with --systems perseus_vault.)