LLM KOSH

LLM KOSH, a local-first memory cartridge for MCP-compatible AI clients, has been released. The tool provides durable, inspectable memory storage without relying on hosted services, featuring a read-only MCP server, CLI, and background service. It is available now for Python 3.10+ with packaging for Windows, macOS, and Linux in progress.

llm-kosh is a local-first memory cartridge for MCP-compatible AI clients. It gives your agents durable memory without handing your workspace to a hosted memory service. Think of it as a structured, inspectable memory layer for agents: - plain files you can back up, diff, and review - a tamper-evident ledger for every mutation - a read-only-by-default MCP server - a background service for intake and maintenance - a CLI for local control and automation - Keep AI context local and auditable. - Separate the cartridge root from the repository root. - Drop receipts or intake files into watched folders and let the service absorb them. - Connect MCP clients with minimal privilege by default. - Publish and verify the same artifact through GitHub Actions. The core project is usable now: - the CLI runs locally - the Python package installs and works - the MCP server runs locally - the service can watch intake folders - the GitHub Actions publish path is working The remaining work is release polish for Windows, macOS, and Linux packaging. Python 3.10 or newer is required. python -m pip install --upgrade llm-kosh llm-kosh install --yes llm-kosh status That installs the package, creates the default cartridge at ~/.llmkosh/cartridge , configures local defaults, and registers the supported desktop integration where possible. To manage the background service: llm-kosh service start llm-kosh service status llm-kosh service stop If you want to work in a custom cartridge location, set the root explicitly: llm-kosh --root ./my-cartridge init --owner "Local User" llm-kosh --root ./my-cartridge add --kind note --title "First memory" --body "Hello" llm-kosh --root ./my-cartridge query "Hello" There are three folders worth knowing: - the repository root: the code checkout you are reading now - the cartridge root: the live memory store selected by --root or LLMKOSH ROOT - watched intake folders: receipts/ , intake/ , and any configured external drop folders If you drop files into the cartridge’s intake areas, the service can process them asynchronously. If you configure external folders through daemon .watched directories , the service can absorb those too. llm-kosh --root ./my-cartridge mcp-server The MCP server starts read-only. Enable stronger capabilities only for clients that should be allowed to write, mutate, or export private context: llm-kosh --root ./my-cartridge mcp-server --allow-write llm-kosh --root ./my-cartridge mcp-server --allow-write --allow-mutate llm-kosh --root ./my-cartridge mcp-server --allow-private You can also run MCP over local HTTP: llm-kosh --root ./my-cartridge mcp-server --http --port 8000 endpoint: http://127.0.0.1:8000/mcp - Python CLI for creating, searching, packing, importing, and verifying cartridges - read-only-by-default MCP server - local background service for intake and maintenance jobs - optional desktop packaging with a bundled CLI sidecar - plain-file storage that stays inspectable, backupable, and Git-friendly - optional extras for filesystem watching, service integration, semantic search, and ingest helpers python -m pip install "llm-kosh watch " filesystem events python -m pip install "llm-kosh server " FastAPI service python -m pip install "llm-kosh semantic " local vector search python -m pip install "llm-kosh ingest " document conversion helpers python -m pip install "llm-kosh all " all optional features MCP support is included in the base installation. python -m pip install -e ". server,watch,ingest " python -m pytest -q If you are changing packaging or release behavior, also run: python -m build python -m twine check dist/ - Storage and search are local by default. - There is no automatic cloud sync or telemetry in the Python package. - MCP starts read-only. - Write, mutation, and private-export capabilities require explicit opt-in. - Context exports are checked for common secret patterns before sharing. - Cartridge files are plaintext; use operating-system disk encryption if local data at rest needs encryption. See SECURITY.md /rastogivaibhav/llm-kosh/blob/master/SECURITY.md and docs/SECURITY.md /rastogivaibhav/llm-kosh/blob/master/docs/SECURITY.md for boundaries and limitations. The Electron desktop app is packaged separately from the Python package. Local developer builds and Windows installer smoke tests are supported. Public GA desktop distribution still requires verified Windows code signing and macOS Developer ID signing/notarization. For the current release posture across package, MCP, service, and desktop, see GA READINESS.md /rastogivaibhav/llm-kosh/blob/master/GA READINESS.md . Quickstart /rastogivaibhav/llm-kosh/blob/master/QUICKSTART.md Architecture /rastogivaibhav/llm-kosh/blob/master/docs/ARCHITECTURE.md CLI reference /rastogivaibhav/llm-kosh/blob/master/docs/CLI REFERENCE.md MCP guide /rastogivaibhav/llm-kosh/blob/master/docs/MCP GUIDE.md Developer guide /rastogivaibhav/llm-kosh/blob/master/docs/DEVELOPER GUIDE.md Developer FAQ /rastogivaibhav/llm-kosh/blob/master/docs/DEVELOPER FAQ.md MCP developer guide /rastogivaibhav/llm-kosh/blob/master/docs/MCP DEVELOPER GUIDE.md Service developer guide /rastogivaibhav/llm-kosh/blob/master/docs/SERVICE DEVELOPER GUIDE.md Desktop developer guide /rastogivaibhav/llm-kosh/blob/master/docs/DESKTOP DEVELOPER GUIDE.md Release engineering /rastogivaibhav/llm-kosh/blob/master/docs/RELEASE ENGINEERING.md Documentation standards /rastogivaibhav/llm-kosh/blob/master/docs/DOCUMENTATION STANDARDS.md GA readiness /rastogivaibhav/llm-kosh/blob/master/GA READINESS.md Archived historical docs /rastogivaibhav/llm-kosh/blob/master/docs/archive/README.md Native C++ math acceleration is optional. Set LLM KOSH BUILD NATIVE=1 and install pybind11 before building if you want to test it. Release wheels use the portable pure-Python fallback. Licensed under the MIT License.