AutoJev
Download

Integration · Agent

Use other models in Kimi Code CLI

Kimi Code CLI is Moonshot AI’s terminal agent and reads providers and models from a TOML config. AutoJev adds one autojev provider plus a model entry for each model or route you select, then routes every request to the provider behind it.

  • Edits ~/.kimi-code/config.toml (or ~/.kimi)
  • One provider, one entry per selection
  • Chat Completions in, any protocol out
  • Original TOML restored on disconnect
Download AutoJevStar on GitHub

Agents

+ Agent

CodexResponses

~/.codex/config.toml

Connected

Claude CodeMessages

~/.claude/settings.json

Connected

Gemini CLIGemini

~/.gemini/settings.json

Connected

OpenCodeChat

~/.config/opencode/opencode.json

Connect

How AutoJev connects Kimi Code CLI

Kimi Code CLI lets you define providers and models in config.toml. On connect, AutoJev adds [providers.autojev] with type openai and the gateway URL http://127.0.0.1:9527/v1, then writes a [models."<id>"] table for every model and route you selected, each pointing at that provider. default_model is set to your chosen default.

Kimi Code CLI then talks Chat Completions to AutoJev, and AutoJev forwards each request to the real upstream — Kimi’s own API through an OpenAI-compatible endpoint, DeepSeek, OpenRouter, a local model or a route that picks between them — converting protocols where needed.

What AutoJev configures

ItemValue
Config file~/.kimi-code/config.toml; an existing ~/.kimi/config.toml is also recognized
Provider[providers.autojev], type = "openai"
Base URLhttp://127.0.0.1:9527/v1
Keyautojev-local-kimi (placeholder; labels requests as kimi)
Modelsone [models."<id>"] per selection, max_context_size = 128000
Defaultdefault_model = your default model or route

What Kimi Code CLI gains from the gateway

One place for keys

Provider keys live in AutoJev on your machine. Kimi Code CLI only holds a placeholder key, and requests go directly to your providers with no markup.

Failover before output

On connection errors or HTTP 401/402/403/408/429/5xx before output starts, a route retries another candidate (up to 4 attempts by default) and respects Retry-After.

Per-agent logs and cost

Every request is labeled kimi, with model, route, input/output/cache tokens, latency and estimated cost. Prompts and responses are not stored.

Routing without edits

Connect a route once. Switch its strategy between load balancing and intelligent selection in AutoJev without touching Kimi Code CLI’s config again.

Set up Kimi Code CLI with AutoJev

  1. 1

    Install AutoJev

    Download the desktop app and open it; the gateway runs on http://127.0.0.1:9527.

  2. 2

    Add providers and models

    Add any OpenAI-compatible endpoint (for example Moonshot’s), DeepSeek, OpenRouter, Ollama or an Anthropic-compatible endpoint with your key, test it and add models.

  3. 3

    Optional: create a route

    Group models into a route such as autojev/fast to let AutoJev choose, balance or fail over between them.

  4. 4

    Connect Kimi

    On Agents, AutoJev detects the kimi executable. Select models and routes, choose the default and click Connect.

  5. 5

    Start Kimi Code CLI

    Start a new session; requests appear in AutoJev Logs under kimi.

    bash
    kimi

The config AutoJev writes

Example with the route autojev/fast as default. Each extra selection adds another [models."<id>"] table.

config.toml
# ~/.kimi-code/config.toml — keys AutoJev writes on connect default_model = "autojev/fast" [providers.autojev] type = "openai" base_url = "http://127.0.0.1:9527/v1" api_key = "autojev-local-kimi" [providers.autojev.custom_headers] x-autojev-agent = "kimi" [models."autojev/fast"] provider = "autojev" model = "autojev/fast" max_context_size = 128000

Switching models and routes

All your selections are present as model entries, so you can switch between them in Kimi Code CLI. To add, remove or change the default, update the Agents page and reconnect, or use “Batch update configurations”. On each update, AutoJev removes the model entries it owns and writes the new set; your own models are untouched. If one of your existing model IDs collides with an AutoJev entry, AutoJev refuses to overwrite it.

Restoring config.toml

AutoJev validates the TOML before writing and keeps a backup. Disconnect, “Stop and restore configurations” or quitting from the tray reverts the default model, provider and model entries it added, while keeping non-conflicting edits you made later.

See every Kimi request

Requests are labeled by agent, so you can filter Logs to kimi and check model, route, tokens, latency and estimated cost.

Logs

TimeModelStatus
14:02:31deepseek/deepseek-chat200
14:02:18openrouter/claude-sonnet200
14:01:55ollama/qwen3:14b200
14:01:40openai/gpt-5-mini429
14:01:12deepseek/deepseek-chat200

Frequently asked questions

Can Kimi Code CLI use non-Kimi models?

Yes. With AutoJev connected, any model you add in AutoJev — DeepSeek, OpenRouter, OpenAI, Ollama or another OpenAI-compatible endpoint — is available as a model entry.

Which config file does AutoJev edit?

~/.kimi-code/config.toml. If that does not exist but ~/.kimi/config.toml does, AutoJev edits the latter.

Does AutoJev have a Kimi/Moonshot preset?

No dedicated preset. Add Moonshot’s API as an OpenAI-compatible provider with its base URL and your key.

Which protocol does Kimi use with AutoJev?

The provider is written with type openai, so Kimi Code CLI sends Chat Completions. AutoJev converts for Responses or Anthropic Messages upstreams.

Will my own Kimi models be removed?

No. AutoJev only replaces model entries whose provider is autojev.

Keep exploring

IntegrationsOpenAI-compatible APIsKimi, Qwen, SiliconFlow, MiniMax, vLLM, LM Studio — any OpenAI-compatible endpoint, available to all your agents.FeaturesOne-click agent setupConnect Codex, Claude Code, Gemini CLI, OpenCode and more. AutoJev edits each native config, backs it up and restores it.FeaturesUsage & request logsTokens, latency and estimated cost for every request, broken down by agent, model and provider. Metadata only, stored locally.FeaturesSmart model routingExpose routes like autojev/fast to every agent. Load-balance models or let Jev pick by cost, quality, speed or balance.Use casesFor multi-agent developersCodex, Claude Code, Gemini CLI, OpenCode and more on one local gateway, with per-agent models and one request log.

Updated 2026-10-14

Give every agent the right model.

Install AutoJev, add the providers you already pay for, and connect your agents in one click. Everything stays on your machine.

Download AutoJevStar on GitHub

Free & open source · macOS · Windows · Linux