LLM conversations are moving from isolated sessions to persistent assistants — that requires memory that is both durable and editable by humans. Basic Memory treats knowledge as plain Markdown files your editor can open and your agent can update, so context accumulates as searchable, linkable notes instead of ephemeral chat excerpts.
What Sets It Apart
-
Local-first, two-way Markdown files. So what? Your notes remain plain text you control (no opaque DB schema), can be edited by humans or agents, exported anytime, and fit naturally into existing editor workflows (Obsidian, VS Code).
-
MCP-native interoperability. So what? Any client that speaks the Model Context Protocol (Claude, Claude Code, Codex, Cursor, Custom GPTs, etc.) can connect to the same memory without bespoke integrations—agents discover tools progressively via behavior hints instead of trial-and-error.
-
Semantic search, schema tooling, and progressive tool discovery. So what? Hybrid full-text + vector ranking finds relevant notes by meaning; schema_infer/validate helps maintain structure across a growing knowledge base; tool annotations let agents safely choose read/write operations.
-
Optional hosted cloud for sync and teams. So what? You can run fully local, air-gapped instances for privacy or opt into the hosted service for cross-device sync, snapshots, and team shared workspaces with built-in backups.
Who it's for — and tradeoffs
Great fit if you want persistent, human-editable context for LLMs and prefer file-first workflows over opaque vector-only stores. It’s well suited for developers, knowledge workers, and teams that want agent-friendly notes that live in your repo or personal folder.
Look elsewhere if you need a purely managed vector DB with turnkey enterprise SLAs, or if you cannot accept the AGPL-3.0 license for local deployments. The local path requires Python tooling for installs; the cloud path is paid (beta pricing noted in the repo).
