The harness with muscle memory.
Keyoku watches how you work in Claude Code, Cursor, or Codex, learns your repeated patterns, and turns them into workflows you run with one command. Approve once — never re-explain a process again.
One command to install. Works out of the box — no API key required. Add a Gemini or Anthropic key when you want model-refined suggestions. Everything stays on your machine.
npx keyoku init|
Activity tracing, pattern detection, workflow execution, goal convergence, connector gateway |
Semantic search, knowledge graph, memory decay, 12-signal heartbeat — powers team sync and the context graph |
Claude Code · Cursor · Codex
│
├─ PostToolUse hook ──► activity log (local JSONL)
│ │
└─ MCP connection ▼
│ pattern detection
│ heuristics find candidates,
│ a small model refines them
│ │
▼ ▼
workflow_execute ◄── approved templates
│
├─ bash steps → run directly (timeouts, audit)
├─ agent steps → hand off to your coding agent
└─ review steps → wait for your sign-off
| Snippets & aliases | Keyoku | |
|---|---|---|
| Discovery | You write them by hand | Learned from how you actually work |
| Steps | Static shell commands | Bash + agent judgment + human review, mixed |
| Execution | Fire and forget | Step-by-step state, pause/resume, full audit trail |
| Scope | One machine, one shell | Any MCP agent; team sync via keyoku-engine |
| Trust | Implicit | Approval is the boundary — nothing runs unsigned |
All data stays on your machine. Open source. Built for agents that do real work.