Markdown source of truth
Use human-readable, version-controllable markdown as agent knowledge.
Agent memory
Persistent, searchable memory for AI agents.
Give agents a local SQLite memory store optimized for markdown, semantic search, hybrid retrieval, and offline-first sync.
SQLite-Memory
SQLite-native extension
Best for
Agent memory and RAG
Repository
GitHubProject
What it is
SQLite-Memory is a SQLite extension that turns markdown content into persistent, searchable agent memory using markdown-aware chunking, vector similarity, FTS5, and embeddings.
Memory databases can stay local for privacy and latency, or sync between agents with SQLite Sync when distributed collaboration is needed.
Why it matters
Persistent and shareable context
for autonomous agents.
Agents need durable context that survives restarts, moves across tasks, and can be queried by meaning. A local SQLite memory file gives that context a simple, inspectable source of truth.
By combining human-readable markdown with semantic retrieval and optional offline-first sync, SQLite-Memory makes agent memory portable, private, and operationally simple.
Capabilities
Features and characteristics
Hybrid retrieval
Combine vector similarity with FTS5 full-text search for better recall.
Markdown-aware chunking
Preserve semantic boundaries when indexing files and documentation.
Local embeddings
Run local embedding models through llama.cpp with no API cost or data egress.
Intelligent sync
Content hashes skip unchanged text, replace modified files atomically, and clean up deleted files.
Transactional safety
Text, file, and directory ingestion use transactions so failed files do not leave partial rows behind.
Sample code
Add knowledge and query memory
Load vector and memory extensions, configure embeddings, ingest content, and search by meaning.
-- Load dependencies and SQLite-Memory
.load ./vector
.load ./memory
SELECT memory_set_model('local', '/models/nomic-embed-text-v1.5.Q8_0.gguf');
SELECT memory_add_text(
'SQLite is a small, fast, self-contained SQL database engine.',
'sqlite-docs'
);
SELECT memory_add_directory('/path/to/docs', 'project-docs');
SELECT path, snippet, ranking
FROM memory_search
WHERE query = 'how do embedded databases store information efficiently';