Five-phase procedure engine
Research, plan, code, verify, PR. Each phase has its own memory tier and tools, so context never leaks between roles.
Post a task in Slack. Memi plans, codes, tests, and ships a PR — with full memory of your codebase and preferences.
Memi listens in Slack, takes tickets from Linear, owns stories in Jira. Same AI engineer — wherever your team talks.
A real engineering pipeline. Memi plans, writes, tests, and ships against the memory of your codebase — not a stateless prompt window.
Research, plan, code, verify, PR. Each phase has its own memory tier and tools, so context never leaks between roles.
Vector search over conventions, decisions, and learnings. Re-reads the things that matter for the phase you're in.
Claude Code or Codex run as subprocess on your API key. Memi orchestrates — the heavy lifting bills to your account.
Per-event token routing means your install never sees another workspace's threads. One bot, isolated by team.
Estimates the task before spawning. Refuses runs that won't fit your remaining balance instead of stalling mid-flight.
Linear or GitHub webhook drops a ticket on Memi. It triages, plans, codes, and ships a PR back — same engine as @memi.
Features, bug fixes, refactors, PR reviews, docs. Memi handles the long tail with full context from your codebase, Slack threads, and connected tools — and a memory that compounds.
Connect Slack. Link your repos. Post @memi and get a PR. No config files, no credit card.