Software Factory

    Your whole Software Factory, run by governed agents.

    Kimchi turns plan, code, review, test, deploy, and monitor into one continuous loop. Agents do the work. Your rules, models, and budgets stay in charge.

    Get KimchiTalk to us
    kimchi factory run KIM-482
    1. ticketKIM-482 Add rate limiting to /v1/chatok
    2. planSpec drafted, 3 files, approved by @danaok
    3. codeAgent on GLM wrote the middleware and configok
    4. test12 tests added, all passingok
    5. reviewPR #918 reviewed: score 92, 1 warning fixedok
    6. deployMerged, released through CIok
    7. monitor$0.84 spent, within team budgetok

    The loop

    One loop, ticket to production and back.

    Pick a stage to see what Kimchi does there.

    ROUTER · OPEN MODELS · POLICIES · BUDGETS · AUDIT TRAIL · SECRETS ·

    01 / Plan

    Tickets become specs

    Agents read the issue, the codebase, and past decisions, then draft a plan your team can approve before a line is written.

    Kimchi Coding

    The Kimchi stack

    Every surface, every agent, one governed platform.

    Work from the CLI, your editor, the browser or the desktop app. Kimchi's gateway, workspaces and inference run as SaaS or in your own infrastructure.

    01 Surfaces
    • CLI
    • VS Code plugin
    • Web IDE
    • Kimchi desktop app
    02 Agents
    Engineering
    • Coding agents
    • Review agents
    • Test agents
    • Your custom agents
    Knowledge work
    • Sales agents
    • Marketing agents
    • Support agents
    • Ops agents
    03 Kimchi platform

    Kimchi

    Gateway
    Control plane
    • Auto-router
    • Policies
    • Budgets
    • Secrets
    • MCP Gateway
    • Audit trail

    Runs in Kimchi cloud, EU region

    Workspaces
    Remote workspaces
    • Persistent cloud dev environments
    • Teleport sessions
    • Remote plan execution
    • Secrets brokering
    • In-region isolation
    • GPU workspacesSoon

    Runs in Kimchi cloud, in your region

    Inference
    Inference
    • GLM, MiniMax, Kimi
    • Dedicated GPUs
    • EU hosting
    • OpenAI-compatible API
    • No training on your data
    • Bring your own providerSoon

    Runs on Kimchi GPUs in Paris

    Each component deploys on its own terms. For example, self-host the gateway and keep inference on Kimchi.

    04 Your tools
    • Git
    • CI / CD
    • Jira and Slack
    • CRM
    • Your VPC

    Capabilities

    What you get, stage by stage.

    From the agent and spend control to the remote, reviewed, secured layer teams need to run agents at scale.

    CapabilityWhat it is
    Work anywhere
    Kimchi agent (harness)Plans, edits and runs code, and handles long multi-step work on any model your admin allows.
    CLIThe full Kimchi agent in your terminal, with live cost and status.
    IDE: VS Code, JetBrains, ZedUse Kimchi in the editor you already have.
    Kimchi StudioDesktop app that creates docs, decks and sheets, and connects to Slack, Atlassian and any MCP server.
    Models and spend control
    Kimchi AI GatewayOne endpoint for Kimchi's open-weight models plus your Anthropic, OpenAI or Copilot accounts.
    Open-weight models hosted in the EUGLM, Kimi, DeepSeek and others on dedicated GPUs in Paris, with no third-party API in the path.
    Auto-routerPicks the best model for each task automatically.
    Budgets and model access controlBudget caps per user, team and org, plus control over which models each team can use.
    Cross-tool spend visibilityAI spend by user, team, model and project across every tool on the gateway.
    Admin policy enforcementEnforce permission modes (for example block yolo mode) across CLI, Studio and IDEs.
    Scale and secure
    Secure, in-region remote workspacesPersistent cloud dev environments for developers and agents.
    Teleport and remote executionMove a session to the cloud, or run an approved plan remotely.
    Code ReviewerAI review on every merge request or pull request.
    MCP GatewayOne governed entry point for every MCP server your agents use, with access policies, secrets and audit logs.
    Secrets brokeringAgents use your credentials without secrets ever being placed in the workspace.
    Cost per PR and ROI reportingProve the return on AI spend, PR by PR.
    PII redactionDetect and redact personal data in prompts before it reaches a model.
    Self-hosted KimchiRun Kimchi in your own cloud or data center, with an air-gapped option.
    Priority supportDedicated support channel and faster response times.
    Coming soon
    Scheduled and event-triggered agentsAgents that start on a schedule or from Jira, GitHub or Slack events.
    GPU workspacesRemote workspaces with GPUs for ML and inference work.
    Bring your own model providerConnect your own OpenAI-compatible inference endpoint.
    Managed MCP connectors in StudioReady-made connectors for common work apps, managed centrally.
    SSO, SCIM and audit logsSingle sign-on, user provisioning and audit trails for enterprise identity.

    Governed by default

    Automation you can actually sign off on.

    Open models

    GLM, MiniMax, Kimi, routed per task.

    Your infrastructure

    Run in your VPC. Your code is never training data.

    Spend limits

    Budgets per team, agent, and run. Hard caps, not alerts.

    Audit trail

    Every agent action and model call recorded.

    Start your software factory.

    Connect a repo, pick your models, and let agents run the loop under your rules.

    Get KimchiTalk to us