What it does
Memory Kernel stores decisions, constraints, tasks, facts and notes in a local SQLite database and retrieves a small set of relevant records using SQLite FTS5 search. It works without a vector database or cloud service. Commands such as search, context and wake-up build compact context packs for an agent prompt instead of loading everything. It is available as an MCP server.
It suits cases where an agent needs earlier decisions, rules or preferences without the full history in the prompt. Use remember to save a known item, or ingest to split raw meeting notes, transcripts or session logs into structured memories. Adding --dry-run previews the result before anything is written, and --interactive offers a guided flow. Records can be inspected, edited or deleted individually.
The command-line tool, memory-kernel, is installed from PyPI or from the repository. Running init creates the local database. Backups are made with export and restored with import. The forget command archives a record out of recall while keeping it restorable, and revise marks an old memory as superseded by a newer one. The decay command archives old, rarely recalled note and fact records based on a retention score; decisions, constraints, tasks and preferences are never decayed. The stats command reports database size and whether the native accelerator is active. A practical guide in Ukrainian is included with the project.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
Memory Kernel
Local-first memory for AI agents: SQLite FTS5, deterministic recall, no vector DB, no cloud.
v0.3.2As shown on registry.modelcontextprotocol.io when we last read it