I’ve been building Codemate, an open-source coding assistant focused on making AI coding workflows more reliable.
Core features:
Multi-agent workflow: planner, research, coder, tester, reviewer, and writer work as separate roles.
TaskGraph planning: complex requests are broken into structured steps before coding starts.
Testing and review loop: implementation, verification, review, retry, and self-check are part of the flow.
Mistake learning: when Codemate fixes an error or discovers a useful pattern, it can save that as a reusable lesson for future runs.
Memory system:
supermemory for user preferences and explicit “remember this” notes
lessons for reusable engineering rules and mistakes to avoid
changelog for recent project history
Drift detection: long tasks are checked against the original goal to avoid going off-track.
Persistence writer: after a run, Codemate records what changed and what should be remembered.
The goal isn’t to replace developers. It’s to make coding agents better at long tasks, refactors, testing, and project maintenance.