Models, providers and configuration

#How keys are set

Autopilot uses one active provider at a time. Every model call goes to that provider with your key: chat, compaction and memory side-calls, embeddings, the model picker, pricing and error messages. Nothing is sent to OpenRouter while another provider is active.

You can provide a key in three ways:

1. Environment variable. Each provider has its own variable (see the table below). With a key in the environment, the setup screen does not appear, which is how you run Autopilot in CI, a VM or a container.
2. Sign-in or setup screen. The first-run screen, /provider inside the UI, or autopilot auth <id> on the command line. The key is checked, then saved to the config file.
3. Config file. providers.<id>.apiKey in config.json. OpenRouter and Requesty keep their own top-level fields (openrouterApiKey, requestyApiKey).

Environment variables win over the file.

#Providers

#Switch provider

autopilot auth <id> accepts these options where they apply:

#Features that need OpenRouter

These switch off on other providers:

• Billed-cost confirmation. Elsewhere, /cost shows estimates from the provider's published prices.
• Subagents in Hotcell sandboxes.
• The Jev completion and message-triage checks.

#Embeddings without an embeddings API

Anthropic, DeepSeek, Moonshot, Kimi Code, the Coding Plan and LM Studio have no embeddings API. Memory still works through text and topic-key recall. To keep vector recall, set "embeddingProvider": "openai" (or AUTOPILOT_EMBEDDING_PROVIDER=openai) with that provider's key, or set memoryEmbeddingModel.

#Local models

#Ollama

Start Ollama, then make it the active provider:

autopilot auth ollama
autopilot

Autopilot checks that it is reachable and discovers your installed models. The default endpoint is http://127.0.0.1:11434/v1. For another host:

autopilot auth ollama --base-url http://gpu-box:11434/v1

Or set it in the config file:

{
  "provider": "ollama",
  "providers": {
    "ollama": { "baseUrl": "http://gpu-box:11434/v1" }
  }
}

#LM Studio

autopilot auth lmstudio

The default endpoint is http://127.0.0.1:1234/v1. Use --base-url for another host.

#Choose a model

Pick a model on first run, or switch with /model. Each provider shows its own catalog: OpenRouter's live catalog, models.dev for other hosted providers, and live discovery for Ollama and LM Studio.

/model                          # open the picker
/model list [filter]            # list models
/model anthropic/claude-sonnet-5

autopilot -m moonshotai/kimi-k2.7-code -p "..."

On OpenRouter, any model id works, including variants such as :free and :nitro. Models listed without tool calling are hidden from the picker. A key with no credits can still use OpenRouter's free models (ids ending in :free), subject to OpenRouter's daily request cap.

#OpenRouter key and credit

/key shows which key is in use. On OpenRouter it also shows what the key has spent and how much credit is left. /key set <key> replaces it and /key clear removes it.

You can also connect from the CLI:

autopilot auth openrouter            # browser sign-in
autopilot auth openrouter --paste    # paste an existing key
autopilot auth openrouter sk-or-...  # save a key directly

Over SSH or in a container, sign-in shows a link to open on any device and asks for the code OpenRouter displays.

#OpenRouter provider routing

OpenRouter serves most models through several upstream providers. Autopilot always requires upstream providers that support every parameter it sends. To steer routing further, add openrouterProvider with any of OpenRouter's provider preferences:

{ "openrouterProvider": { "ignore": ["SomeProvider"], "data_collection": "deny" } }

Avoid order and sort unless you need them. They turn off sticky routing, which keeps the prompt cache warm.

#Custom gateway endpoint

Point every model call at your own OpenAI-compatible endpoint:

export KIMIFLARE_BASE_URL="https://your-broker.example.com/v1"  # /chat/completions is appended
export KIMIFLARE_API_KEY="<bearer for that endpoint>"           # optional
autopilot -p "..."

The same pair can be saved in the config file as baseUrl and apiKey. Environment variables win, field by field. With a base URL set:

• Chat requests go to <baseUrl>/chat/completions and memory embeddings to <baseUrl>/embeddings.
• Authorization: Bearer $KIMIFLARE_API_KEY is sent. With no key set, no Authorization header is sent.
• Model ids pass through unchanged. Your endpoint owns provider dispatch.
• Cost is a local estimate only.

#Requesty

Requesty is another OpenAI-compatible gateway. It is only used when no OpenRouter key and no custom endpoint are configured.

export REQUESTY_API_KEY="<your key>"
autopilot                 # or: autopilot auth requesty

• The key can also be saved as requestyApiKey in the config file. The environment variable wins.
• REQUESTY_BASE_URL picks a region, for example https://router.eu.requesty.ai/v1. Only https URLs on Requesty's hosts are accepted.
• Defaults on Requesty: kimi-k2.6 for the main model, deepseek-v4-flash for side calls, and openai/text-embedding-3-small for memory embeddings.

#Config file

The config file is ~/.config/kimiflare/config.json. If XDG_CONFIG_HOME is set, it is $XDG_CONFIG_HOME/kimiflare/config.json. Autopilot writes it with mode 600. The kimiflare path is kept for compatibility with earlier versions.

{
  "provider": "deepseek",
  "providers": {
    "deepseek": { "apiKey": "sk-...", "model": "deepseek-v4-pro" },
    "ollama": { "baseUrl": "http://gpu-box:11434/v1" }
  }
}

Hooks are configured in separate settings.json files. See Hooks.

#Config keys

The full type, with comments, is KimiConfig in src/config.ts.

#MCP servers

MCP servers are defined under mcpServers. Each entry is either a local command or a remote URL:

Inside the UI, /mcp list shows connected servers and their tools, and /mcp reload reconnects them. MCP tool names start with mcp_.

#Language servers

Turn on LSP with lspEnabled and define servers under lspServers, or use /lsp config in the UI. Each server takes command (string array), and optionally env, enabled, rootPatterns, timeoutMs (default 10000) and maxRestartAttempts (default 3). A project can ship its own servers in .kimiflare/lsp.json; they only start after you trust the project.