Memory before repeated grep
Claude Code can search code for current symbols, but past decisions and preferences often live outside the current source tree. Palace gives those memories a reliable retrieval path.
Claude Code memory
Persistent memory for Claude Code
Claude Code
Palace gives Claude Code durable recall for prior fixes, repo conventions, and team decisions through MCP.
Works with Cursor, Claude Code, Codex, and other MCP-capable clients.
Combines semantic retrieval, BM25-style search, diaries, and knowledge graph recall.
Open-source MIT core for solo developers; paid server for shared team memory.
Palace vs built-in vendor memory
| Feature | Palace | Built-in memory |
|---|---|---|
| Cross-agent | Cursor, Claude Code, Codex, Claude Desktop | Usually one vendor product |
| MCP native | Yes | Varies; often not a shared MCP layer |
| Team memory | Shared self-hosted server | Often personal / per-account |
| Self-hosted | Yes — your infra | Typically vendor-hosted |
| Provenance | Wing / room / source citations | Often opaque personalization |
| Agent workflow | Explicit memory protocol for agents | Convenience memories inside one tool |
Beyond session and project memory alone
Vendor memory inside one tool is convenient. Palace is for engineering teams that need shared, source-grounded recall across agents — and a self-hosted option when cloud memory is a non-starter.
Shared team layer
With Palace Server, one developer's hard-won setup or debugging lesson can become available to the next agent session on the team.
Private deployment
Team memories stay in your server and database. Local embeddings keep sensitive context out of external embedding APIs.