What it does
myc is a local task-and-memory layer for coding agents. It combines a task queue, a log of facts and decisions, search over a project's memory that mixes lexical and vector matching, and a built-in code index. The README says it makes no network calls of its own and needs no mandatory LLM key. Two commands set a project up: myc init creates a workspace with a SQLite database in the repository, and myc wire adds hooks and MCP configuration for agents such as Claude Code, Codex, opencode and Kimi. Agents then claim tasks, record decisions with remember, retrieve them with recall, and receive a context packet at session start from prime. Knowledge can be anchored to spans of code and follows the code as it moves; when the code is gone, the entry ranks lower instead of being deleted. The code index uses tree-sitter grammars fetched on demand and answers symbol, caller and code search queries. A run command makes heavy test and build commands from several agents on one machine take turns. It installs with npm, which brings a native binary for macOS, Linux and Windows (Windows support is described as experimental), with Bun as the fallback engine. The project is MIT licensed and includes an import path from the beads task tracker.
Capabilities
- Task queue and decision log that survive context compaction
- Hybrid lexical and vector search over project memory
- Knowledge anchored to code and marked down when the code moves
- Built-in tree-sitter code index for symbols, callers and code search
- Wires hooks and MCP for Claude Code, Codex, opencode and Kimi
- Queue that makes heavy test and build runs take turns
When to use it
- Your coding agents lose decisions when their context is compacted
- You run several agents on one machine and want builds and tests to queue
- You want agent memory kept in a local SQLite database rather than a hosted service
