Tracing that's automatic
A PostToolUse hook records every command, edit, and tool call your agent makes. Entities extracted, stored as local JSONL. No cloud, no telemetry, nothing to configure.
KeyokuOpen source · MCP-native · Local-first
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.
Automatic activity tracingHeuristic + model pattern miningOne-command workflows
Agent ChatLive
Wire Keyoku into Claude Code in one command — activity hook, MCP server registration, local state. It starts learning from your very next session.
1
Wire the hook + MCP server
npx keyoku init2
Restart Claude Code — tracing is live
3
After a few sessions, ask your agent to run
workflow_suggestPatterns that surface
Non-overlapping sequence mining finds your real workflows — deploy routines, test-fix loops, PR prep. An optional small model filters noise, names drafts, and parameterizes commands.
Autonomy you control
Bash steps run directly. Judgment steps hand off to your coding agent. Human-review steps wait for you. Approval is the trust boundary — nothing runs that you didn't sign off.
How it worksSimple outside.
Simple outside.
Structured underneath.
Keyoku turns your real working patterns into runnable workflows, then keeps you in control of every action.
Observe
Record every command, edit, and tool call as structured activity.
A PostToolUse hook captures your agent's real work — entities extracted, stored as local JSONL. No cloud, no telemetry, zero configuration.
Learn
Mine repeated sequences into draft workflows.
Non-overlapping pattern detection finds real routines and collapses them into the longest repeated chain. An optional small model filters noise, names drafts, and parameterizes commands.
Run
Execute approved workflows with one command — judgment included.
Bash steps run directly with timeouts and a full audit trail. Judgment steps hand off to your coding agent and resume on completion. Human-review steps wait for explicit sign-off.
Get started
Pick your path.
Recommended
Claude Code
Wire the activity hook and MCP server in one command. Keyoku starts learning from your very next session.
Get started →Any MCP agent
Cursor & Codex
Native MCP integration — the same tools and workflows in any MCP client. The event schema is agent-neutral by design.
View tools →CLI
Inspect status, mine patterns, review approvals, and audit every action keyoku takes — straight from your terminal.
View CLI →Open Source
MIT-licensed TypeScript, 200 tests, documented security model. Read the code and audit the trust boundary yourself.
View source →Stop re-explaining your workflows.
Approve once. Run with one command. Start in ten seconds.