ContextPool

ContextPool provides a persistent memory layer for AI coding agents, allowing them to retain engineering insights across sessions. After installation, a single binary scans prior interactions with tools such as Claude Code, Cursor, Windsurf, and Kiro, using an LLM to distill actionable knowledge like bug root causes, fixes, design decisions, and common gotchas. The extracted summaries are stored and automatically recalled at the start of a new session via the MCP protocol, eliminating the need for manual prompting.

The system is designed for developers who work with AI assistants and need continuity in debugging and decision‑making. It works on macOS, Linux, and Windows without additional runtime dependencies, and integrates with existing authentication for Claude Code. Stable project identifiers derived from git remote URLs ensure consistent references across teammates, while multi‑backend LLM routing provides resilience by falling back among Claude, Anthropic, OpenAI, and NVIDIA services.

Team memory is supported through shared pools, where insights pushed by one member become available to others, and API keys are securely stored in the system keychain with a file fallback. A free local mode is offered, with optional cloud synchronization for collaborative use.