TinyContext

Context that fits your local LLMs.

Part of TinySuite. Comes built in to TinyWebUI, and works with any MCP client.

Read the documentation → tinysuite.dev/docs/tinycontext

TinyContext is the memory layer of TinySuite, a set of small, local-first tools that make agents spend tokens on useful work rather than overhead. It stores concise memories and their embeddings in SQLite, ranks them with hybrid BM25 and dense retrieval, and returns only the context that fits the requested token budget, so an agent recalls a small, ranked set of memories instead of resending its history.

No hosted account. No giant context dumps. No required vector database. Against an agent that resends every stored memory, recall used 96.7% fewer tokens, with p95 latency around 12 ms at 20,000 memories (docs/benchmarks.md).

Highlights

• No vendor in the loop. No account, no required API key, no per-request billing, no hosted memory service. Memories stay in a SQLite file you own.
• Recall that fits the budget. Ranking and token trimming happen locally, so only the memories that fit reach the model.
• No required vector database. Embeddings and BM25 live in the same SQLite row; sqlite-vec handles similarity.
• Profile memories. Durable facts such as what to call the user ride along with every recall, under their own small budget.
• Six small MCP tools. Save, recall (by meaning or most recent first), list, get, update and delete, with short stable refs for corrections.
• Self-maintaining. Near-duplicates are skipped at save time, and changing the embedding model re-embeds the store in the background.
• Multi-user when hosted. Behind a trusted proxy, each user gets a separate SQLite store, checked on every message.
• Works where agents already work. MCP over stdio or Streamable HTTP, Python, FastAPI, or Docker. Local ONNX embeddings are the default.

Quick start

... View full README on GitHub