agentforge-memory-sqlite

agentforge-memory-sqlite

SQLite-backed MemoryStore and VectorStore for AgentForge.

Zero external services required — the database lives in a single file (or :memory: for tests). Suitable for development, single-host deployments, and small-to-medium RAG corpora (~10k vectors).

Install

uv add agentforge-memory-sqlite

Usage

from agentforge_memory_sqlite import SqliteMemoryStore, SqliteVectorStore

# Claim audit log
async with SqliteMemoryStore.from_path("agent.db") as memory:
    await memory.put(claim)

# Semantic search
async with SqliteVectorStore.from_path("agent.db", dimensions=1024) as store:
    await store.upsert(items)
    matches = await store.search(query_vector, limit=5)

Both classes pass agentforge_core.testing.run_memory_conformance / run_vector_conformance so they're drop-in replacements for the in-memory defaults.

Performance

SqliteVectorStore does brute-force cosine search in Python: O(N) per query. Fine for ~10k vectors; sluggish past that. v0.2 will add an opt-in sqlite-vec extension path declared via the "native_ann" capability flag.