Skip to content

Releases: TinySuiteHQ/TinyContext

Release list

v0.2.0

v0.2.0 Latest
Latest

Choose a tag to compare

View all tags
@MarcellM01 MarcellM01 released this 31 Jul 14:58
v0.2.0
b08d37e

Highlights

  • Added hybrid BM25 + dense retrieval using SQLite FTS5 and sqlite-vec.
  • Added structured recall context, relevance labels, and configurable RRF cutoffs.
  • Added benchmark results and methodology to the README.
  • Hardened Docker execution, dependency scanning, and release image configuration.
  • Fixed SQLite connection shutdown synchronization for background reindexing, preventing Python 3.14 CI crashes.

Validation

  • 54 tests passing.
  • Source distribution and wheel built successfully.
Assets 4
Loading

TinyContext v0.1.0

TinyContext v0.1.0

Choose a tag to compare

View all tags
@MarcellM01 MarcellM01 released this 31 Jul 08:03
v0.1.0
3861739

TinyContext v0.1.0

The first public release of TinyContext, a token-light local memory layer for AI agents.

TinyContext stores concise memories and embeddings in SQLite, combines lexical and semantic retrieval, and returns only the context that fits the requested token budget.

Highlights

  • Local-first memory storage with SQLite
  • Hybrid BM25 and dense vector retrieval
  • Local ONNX embeddings with reusable model bundles
  • Weighted reciprocal-rank fusion and token-budget trimming
  • Python API for saving and recalling memories
  • MCP support over stdio, SSE, and Streamable HTTP
  • Optional FastAPI server
  • Persistent multi-architecture Docker deployment
  • Automatic migration and embedding backfill for existing databases
  • Portable cosine-ranking fallback for macOS Python builds that cannot load SQLite extensions
  • Python 3.12, 3.13, and 3.14 support across Linux, macOS, and Windows

Install

Core Python library:

pip install tinysuite-context

MCP and server support:

pip install "tinysuite-context[server]"

Run the MCP server directly:

uvx --python 3.12 --from "tinysuite-context[server]" tinycontext

Docker:

docker pull marcellm01/tinycontext:v0.1.0

Community

A special thank you to @benmaster82 for contributing PR #3. Ben replaced the per-call SQLite connection pattern with a thread-safe connection pool, enabled WAL mode, added a busy timeout, and added explicit connection cleanup. This makes concurrent FastAPI and MCP requests more reliable and prevents common database is locked failures.

Notes

  • The selected ONNX embedding bundle is downloaded on first use and reused locally.
  • Memory databases and downloaded models remain local.
  • Existing TinyContext databases are upgraded in place.
  • No hosted account or external vector database is required.
Loading