coree
Persistent memory and code intelligence for AI agents. Runs locally. Works across sessions.
# add to marketplace then install claude plugin marketplace add github:coree-ai/coree claude plugin install coree
What it does
Persistent memory
Stores typed memories - decisions, gotchas, how-it-works notes - and surfaces them at session start. Agents remember what they learned yesterday.
Hybrid search
One search(query) call searches both memories and source code simultaneously. Vector similarity plus BM25 keyword matching, ranked together.
Code intelligence
Tree-sitter parses 20+ languages into an indexed database. Git history is indexed for churn analysis and hotspot detection. get_symbol(name) returns signature, history, and volatility score.
Multi-agent
One memory store, multiple agents. Switch between Claude Code, Gemini, Codex, and Cursor - the context follows you.
Local-first
SQLite by default. No account required, no data leaves your machine. Optional Turso remote sync for cross-machine access.
Zero config
Works immediately after install. Add a .coree.toml to override storage backend, remote sync, or indexing behaviour.
Supported agents
How it works
Session start
The agent calls session_context(). Pending notes from previous sessions are reviewed and promoted to memories. Prior decisions and discoveries are surfaced.
During work
The agent calls search(query) before touching unfamiliar code. Stores findings with store_memories() as it goes - decisions, gotchas, how-it-works.
Code lookups
get_symbol(name) returns a function's signature, its git line-range history, commit churn count, and hotspot score without reading the file directly.
Next session
Context is already there. The agent does not have to re-read the codebase or re-discover architectural decisions. Accumulated knowledge compounds over time.