evo

Local CLI coding agents are stateless by design: every new session starts from zero, and the preferences you have corrected repeatedly, the project conventions you only nailed down after three stumbles, all vanish the moment the terminal closes. evo tackles this with a recall-and-reflect loop — before each model step it pulls relevant memories into the context window, and after each successful turn it distills the conversation into structured memory items, so the agent accumulates project knowledge the way an experienced colleague would.

Every memory carries scope, kind (fact, constraint, procedure, skill), tags, and source metadata. Procedural SOPs live as first-class skills with lessons-learned tracking, separate from declarative facts. All data stays local in SQLite by default; the only outbound calls are reflection model requests. evo is provider-neutral, working with DeepSeek Harness, Claude Code, and Codex through adapters, and it auto-imports existing workspace files such as CLAUDE.md, AGENTS.md, and project directories.

If you work daily with a local coding agent and are tired of re-explaining context every session, evo is worth adding to your stack. You do not need to hand-write memory — the agent distills its own experience as it works.