The governed
    AI coding platform

    Give engineers autonomous coding agents without losing control of models, data, quality, or spend. Start serverless. Deploy in your VPC when ready.

    $curl -fsSL https://github.com/getkimchi/kimchi/releases/latest/download/install.sh | bash
    See Enterprise Governance
    No card required
    kimchi
    Kimchi
    v0.0.26
    ~/project · main
    ›Try a /ferment workflow
    The agent breaks the task into milestones, self-evaluates its output and delivers with minimal interruptions.
    ›Skip
    ↑↓ navigate · enter selectkimchi agent
    Kimchi CLI workflow picker
    Works with
    OpenCodeClaude CodeCursorOpenClawGSD2
    The coding agent

    Prompt to PR. 
    No babysitting required.

    OSS MODELS

    Built for OSS models.

    • 50+ models. Zero lock-in.
    • Automatically orchestrates models by task type and cost
    • Frontier OSS models via our Minimax partnership
    kimi-k2.7reasoning262k context$4.50/1M output tokens
    minimax-m3execution1M context$1.20/1M output tokens
    nemotron-3-super-fp4execution1M context$0.75/1M output tokens
    Avg. price of proprietary models$12.10/1M output tokens
    /FERMENT WORKFLOW

    Hand off the task.
    Come back to a finished PR.

    A workflow that allows you to plan upfront, approve once and forget the terminal. The agent builds through each milestone autonomously — a PR appears when it's done, not a question.

    /ferment implement a "to do" appDon't ask permissions
    ✓  plan approved
    ✓  Phase 1: Scaffold project structure0.8s
    ✓  Phase 2: Build task model + local storage1.1s
    •  Phase 3: Create UI componentsrunning…
    •  Phase 4: Add due dates · tests · open PR
    Phase 3/4: "Create UI components" · step 1/2 · Stop: When done
    CODE SESSION

    Stay in the loop. Have more control.
    Make it quick.

    The standard coding agent conversation powered by our open source model orchestration. Interact more closely with the agent and stir it exactly in the direction you want.

    Rename userId to accountIdAuto permissions
    Found 6 more references in other files.
    Want me to update those too?
    ❯ Yes
      No
    ONE COMMAND SETUP

    Migrate from Claude Code. Keep everything.

    On first run, kimchi setup detects existing Claude Code and OpenCode installations MCP servers, skills, configuration and offers to migrate in one prompt.

    ┌  Claude Code configuration found
    │  MCP servers: filesystem, github, ripgrep
    │  Claude Code skills: 4 in ~/.claude/skills
    │  OpenCode skills: 2 in ~/.config/opencode/skills
    ◇  Migrate MCP servers to Kimchi?
    ● Migrate now
    ○ Skip this time
    BUILT IN MODEL ORCHESTRATION

    Right model per task. Automatic.

    The orchestrator classifies each task and routes to the best model specialized sub-agents run security audits, tests, and analysis in isolated parallel contexts. Turn multi-model off to lock a single provider.

    multi-model
    REASONING / PLAN
    kimi-k2.5
    orchestrator12%
    EXECUTION
    minimax-2.7
    subagents · parallel88%
    EVALUATOR

    Every task is graded before it ships.

    After each milestone, a high-reasoning model scores the output against the original intent catching regressions, half-done work, and silent failures before they compound. Use Opus, Kimi, or whatever model you trust most for the job.

    evaluator · claude-opus-4-7milestone 3 of 4
    spec match9.2 / 10
    regressions0 found
    edge cases2 flagged
    ✓ milestone passed · proceeding to milestone 4
    /TELEPORT

    Hand off to a remote workspace.

    Run /teleport to provision a persistent sandbox pre-loaded with your tools, skills, and agent state. Close the lid — the session keeps running. Reconnect via /sessions or VS Code Remote SSH.

    /teleport auth-refactorpersistent
    ✓  Workspace ready · ws-f3a1b2c
    ✓  Synced 1,204 files
    ✓  Skills and tools loaded
    Session running remotely. Ctrl+D to detach · /sessions to reconnect.
    Coming soon
    Kimchi Studio

    All the power of the CLI harness but inside a visual kanban UI. Chat with an agent to set up tasks, then watch them run in parallel.

    Kanban board illustration
    Live
    Remote Agents

    Spin up coding agents that run remotely in the cloud. Kick off long tasks from anywhere and pick up the results when they're done — no local machine required.

    Remote cloud agent illustration
    Login & try ->
    Governance that enforces your policies

    Spend limits for
    total ease of mind.

    See exactly who, what, and where your AI spend is going. Per-developer, per-team, per-model, per-tag - in real time. Hard budget caps at every level: user, team, API key, org. Caps that enforce when you configure them, not alerts that just notify when it's too late.

    • Per API key set limits for CI/CD pipelines, bots, or contractors
    • Per user every developer has their own budget with real-time visibility
    • Per team finance gets a clean number; engineering keeps shipping
    • Per org hard ceiling across all keys, users, and teams so total spend never exceeds your number
    Budget controls
    APR 2026
    ORG · ACME INC
    $4,820 / $6,000 cap
    80%
    TEAM · PLATFORM ENG
    $284 / $400 limit
    71%
    PER DEVELOPER
    ravi-sharma$38.20 / $50
    maria-t$29.40 / $50
    ci-pipeline-key$12.10 / $100
    ⚠
    ravi-sharma hits 80% budget tomorrow at current run rate.
    Akamai
    BMW Group
    Cisco
    FICO
    Hugging Face
    Kakao
    PlayPlay
    Roku
    ShareChat
    Swisscom
    WeWork
    Akamai
    BMW Group
    Cisco
    FICO
    Hugging Face
    Kakao
    PlayPlay
    Roku
    ShareChat
    Swisscom
    WeWork
    "Kimchi is fundamentally changing how our engineering team thinks about AI-assisted development. The multi-model approach keeps the quality of a single frontier model while cutting token costs dramatically."
    Dekel Shavit · Senior Director of Engineering, Akamai

    Build with freedom.
    Start in 60 seconds.

    Install the CLI, run /ferment, and hand off to the agent. No lock-in, no waiting list, no credit card.

    ★ Star on GitHub
    • Optimized for open-source models
    • No rate limits or token caps
    • Built in model orchestration
    • Open source CLI