ContextPool

ContextPool solves the "amnesia" problem for AI coding agents by providing persistent memory across sessions. It scans past Cursor and Claude Code sessions, extracts actionable engineering insights using an LLM, and makes them available to your agent at session start via the MCP protocol. Local mode is free forever, with optional cloud sync for team-shared memory pools.

• Persistent Memory — Install the CLI, run cxp init, and your agent automatically loads relevant past context at every session start — no manual prompting required.
• Session Scanning — CXP scans your past Cursor and Claude Code sessions and distills bugs, fixes, design decisions, and gotchas into structured engineering knowledge.
• MCP Integration — Agents query memory via standard MCP tools with zero custom integration glue; Claude Code requires zero config, while other IDEs need a single JSON entry.
• Multi-Backend LLM Routing — Provider fallback chain (Claude CLI → Anthropic API → OpenAI → NVIDIA) ensures resilient insight extraction even if one provider is unavailable.
• Team Memory Sync — Push insights to a shared cloud pool so teammates' agents benefit from collective debugging and decision history; project IDs are derived from git remote URLs for consistent resolution.
• Privacy-First Design — Raw transcripts never leave your machine; secrets are redacted before LLM processing and before any cloud sync; only extracted insights are optionally synced.
• Keychain Storage — API keys are stored in the system keychain with a safe file fallback, keeping credentials secure across all supported platforms.
• Broad IDE Support — Works with Claude Code, Cursor, Windsurf, and Kiro with minimal configuration.