The team problem
Every developer teaches their agent the same conventions, build quirks, and architectural decisions. Without shared memory, that context resets over and over — and disappears when someone switches tools.
Shared AI memory for teams
Shared AI memory for engineering teams
team recall
Palace Server turns agent memory into a shared engineering asset: searchable across Cursor, Claude Code, and Codex, hosted on your infrastructure.
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.
Shared team memory vs siloed per-tool memory
| Feature | Palace Server | Per-tool memory |
|---|---|---|
| Shared across the team | Yes — one memory layer | Usually personal or tool-siloed |
| Cross-agent | Cursor + Claude Code + Codex | Resets when tools change |
| Data ownership | Your Postgres / infra | Vendor cloud by default |
| MCP native | Yes | Rarely a shared MCP backend |
| Pricing model | Flat team tiers | Often bundled or opaque |
| Adoption path | Local free core → team server | Start over per product |
What shared memory means here
Palace Server gives the team one MCP memory layer with provenance: wings, rooms, diaries, and knowledge-graph facts that future agent sessions can search.
Choose Palace when
You need shared recall for coding agents, flat team pricing, and ownership of the database and embeddings — not another cloud memory silo per tool.
How teams buy
Developers prove the workflow with the free local core. When shared recall matters, deploy Palace Server and attach a team license.