Oh My Kimi
Verified agent runtime for Kimi Code. Stable daily-use core with alpha orchestration surfaces; release-gated and evidence-gated workflows. | Kimi Code向け検証済みエージェント実行基盤。日常利用コアは安定、オーケストレーション面はalpha。
Last verified:
What is Oh My Kimi?
Oh My Kimi (Oh My Kimi) is a production-ready multi-agent orchestration harness for Kimi Code CLI (K2.6). It is not a prompt pack or model buffet, but a Kimi-native control plane designed for shipping code with verification. Oh My Kimi coordinates, verifies, remembers, and guards the coding process while Kimi writes the actual code.
Oh My Kimi solves six hardest problems in AI-assisted coding: premature completion, context corruption, memory loss, operator blindness, model chaos, and runtime drift. Its core features include Evidence Gates (agents can't claim 'done' without proof like files, diffs, summaries, or passing commands), Worktree Isolation (parallel workers run in isolated Git worktrees for reversible experiments), Local Graph Memory (durable project memory storing goals, decisions, risks, tasks, commands, files, evidence, and concepts as a graph), HUD & Cockpit (live operator visibility into run state, TODOs, ETA, usage, workers, and changed files), DAG Execution (requests become task graphs with roles, dependencies, retries, fallback routing, and evidence requirements), and Decision Trace (every policy decision recorded in decisions.jsonl for inspectable and reproducible runs).
The tool is organized around five operating rituals: Ship (implement with verification via omk chat, omk parallel, omk verify), Inspect (run history & current state via omk runs, omk replay, omk inspect), Design (visual/product direction via omk design, omk open-design), Remember (durable project context via omk graph view, omk index), and Guard (safety & release confidence via omk doctor, npm run release:check). It is designed for developers using Kimi Code CLI who want structured, spec-driven coding teams with verification.
Oh My Kimi is built on 12 architectural pillars with four defining the runtime: Kimi-native Control Plane (Kimi remains primary writer), Evidence-gated Verification (completion requires evidence, no narration-only completion), Worktree-isolated Parallel Lan
Oh My Kimi pricing
Pricing model: Freemium
Free and open-source. Available via npm package @oh-my-kimi/cli with global installation via 'npm install -g @oh-my-kimi/cli'. Version v1.1.12 stable. No paid tiers mentioned - the tool is a free CLI orchestrator for Kimi Code CLI.
Oh My Kimi pros
- Evidence Gates prevent premature completion without proof
- Worktree Isolation makes experiments reversible
- Local Graph Memory provides durable project context
- HUD & Cockpit give live operator visibility into run state
- DAG Execution enables structured task graphs with dependencies
- Decision Trace in decisions.jsonl makes runs inspectable
- Kimi-native design keeps Kimi as primary code writer
- No narration-only completion - requires actual files and diffs
- Parallel workers run in isolated Git worktrees
- Five operating rituals organize daily work systematically
- Automatic loading of project rules, skills, hooks, MCP status
- Stable commands provide normal operator path for daily use
- Quick 10-second installation via npm
- Built-in omk doctor checks system health
- Git diff validation ensures non-empty changes before completion
Oh My Kimi cons
- Requires Node.js 20+ which may be outdated for some systems
- Requires tmux for team/HUD workflows on Unix-like systems
- Depends on Kimi Code CLI being installed and authenticated
- Some commands like omk parallel are still Alpha/Experimental
- Not model-agnostic - intentionally Kimi-first design
- Git worktree isolation requires Git to be installed
- Learning curve for five operating rituals and 20+ commands
- May be overkill for simple single-prompt coding tasks
Frequently asked questions about Oh My Kimi
What is Oh-my-Kimi?
Oh-my-Kimi (OMK) is a production-ready multi-agent orchestration harness for Kimi Code CLI (K2.6). It is not a prompt pack or model buffet, but a Kimi-native control plane for shipping code with verification. OMK coordinates, verifies, remembers, and guards the coding process.
What problems does OMK solve?
OMK solves the six hardest problems in AI-assisted coding: premature completion, context corruption, memory loss, operator blindness, model chaos, and runtime drift.
How do Evidence Gates work?
Evidence Gates require agents to provide proof before claiming 'done'. Completion requires files, diffs, summaries, or passing commands. No narration-only completion is accepted - the system checks required files, diffs, and summaries before marking completion.
What is Worktree Isolation?
Worktree Isolation means parallel workers run in isolated Git worktrees. Experiments stay reversible, and merge is always a deliberate step rather than a side effect. This keeps experiments contained until review/merge is intentional.
How does Local Graph Memory work?
Local Graph Memory stores durable project memory as a graph containing goals, decisions, risks, tasks, commands, files, evidence, and concepts. This provides smaller, safer context for Kimi while maintaining persistent project context across sessions.
What are the five operating rituals?
The five operating rituals are: Ship (implement with verification via omk chat, omk parallel, omk verify), Inspect (run history & current state via omk runs, omk replay, omk inspect), Design (visual/product direction via omk design, omk open-design), Remember (durable project context via omk graph view, omk index), and Guard (safety & release confidence via omk doctor, npm run release:check).
What are the system requirements?
OMK requires Node.js 20+, Git, Kimi Code CLI installed and authenticated, and tmux for team/HUD workflows on Unix-like systems. Installation is done via 'npm install -g @oh-my-kimi/cli'.
Which CLI commands are stable?
Stable/daily-use commands include: omk cockpit, omk design, omk lsp, and omk runs. Start Here commands include: omk menu, omk init, omk doctor, omk chat, omk plan, omk hud. Advanced commands include omk graph, omk mcp, omk replay, omk inspect, omk diff-runs. Alpha/experimental commands include omk parallel, omk run, omk verify, omk goal, omk team.
How is OMK different from other harnesses?
OMK is intentionally Kimi-first. While other models can advise, review, or QA, the run remains bounded. OMK is for Kimi Code with verification, featuring a Kimi-native DAG runtime with evidence gates and graph memory. Other harnesses include OMC for Claude Code users, OMX for Codex CLI users, and OMO for open multi-model routing.
How do I get started with OMK?
From zero to a verified run in six commands: 1) npm install -g @oh-my-kimi/cli, 2) omk --version, 3) omk doctor, 4) omk init, 5) omk plan 'Add a settings page with tests', 6) omk parallel 'Implement the settings page', 7) omk verify --json, 8) omk summary-show. OMK automatically loads project rules, skills, hooks, MCP status, and Git state.