· · · EST. 2026 · · ·
Palethra is a persistent, cross-session memory service for AI assistants. Your agent walks the halls, files what matters into drawers, and remembers — across every conversation, forever.
$ pip install palethra $ palethra setup $ claude mcp add palethra -- palethra-mcp › claude now has a memory.
Works with Claude Code · Claude Desktop · any MCP client
I.
Palethra borrows its shape from the classical art of memory. Every thought your agent keeps lives somewhere — in a wing, in a room, in a drawer it can open again at will.
Top-level domains — projects, people, research threads. Your agent files each memory under the wing it belongs to, so recall is focused rather than flat.
Aspects within a wing — meetings, decisions, bugs, open questions. Rooms are named with hyphenated slugs so your agent can walk directly to them.
The notes themselves — verbatim, never paraphrased. Semantic search across 384-dimensional embeddings returns the exact words, not a reconstruction.
II.
Facts have beginnings and ends. Your agent learns that Max switched from swimming to chess without losing the history of either.
Each assistant gets its own chronicle. Yesterday's Claude leaves notes for tomorrow's Claude in a language they both read fluently.
Postgres row-level security means your memories cannot leak into another user's palace. Not by policy — by physics.
One command to install, one line to register with Claude. No new mental model — the tools are the ones your agent already knows.
III.
Start free. Keep your memories as long as you like.
For the curious.
For the long game.
For whole teams.
IV.
Bugs, questions, feature requests, strong opinions — all welcome.