Providers

KodaCode separates provider identity from provider credentials.

• config.yaml declares which providers exist and which model routes to use.
• auth.yaml stores API keys and OAuth state.
• Model references use provider_id/model_id, such as openai/gpt-5.

The fastest setup path is the /connect dialog inside the TUI. It writes the provider entry and credential entry for you. You can also edit the files manually when you want repeatable machine setup.

Files

Default locations:

File	Purpose
~/.config/kodacode/config.yaml	Provider IDs, base URLs, model routes, budgets, and runtime settings
~/.config/kodacode/auth.yaml	API keys, OAuth tokens, and provider account state

Credentials do not belong in config.yaml.

Built-in providers

The /connect dialog includes native providers and OpenAI-compatible presets:

Provider ID	Name	Notes
openai	OpenAI	OpenAI Platform API key
openai-codex	OpenAI Codex OAuth	ChatGPT/Codex OAuth account created by the OpenAI /connect flow
anthropic	Anthropic	Native Anthropic Messages API
google	Google	Native Gemini API
nvidia	NVIDIA	NVIDIA API endpoint
github-copilot	GitHub Copilot	OAuth/token-backed GitHub Copilot route
deepseek	DeepSeek	DeepSeek API route
qwencloud	QwenCloud	OpenAI-compatible preset
openrouter	OpenRouter	OpenAI-compatible preset
togetherai	Together AI	OpenAI-compatible preset
groq	Groq	OpenAI-compatible preset
fireworks-ai	Fireworks AI	OpenAI-compatible preset
mistral	Mistral	OpenAI-compatible preset
cerebras	Cerebras	OpenAI-compatible preset
deepinfra	Deep Infra	OpenAI-compatible preset
moonshotai	Moonshot AI (Kimi)	OpenAI-compatible preset
venice	Venice AI	OpenAI-compatible preset
zai-coding-plan	Z.AI	OpenAI-compatible preset
ollama-cloud	Ollama Cloud	OpenAI-compatible preset
ollama	Ollama	Local OpenAI-compatible preset; API key is optional
lmstudio	LM Studio	Local OpenAI-compatible preset; API key is optional
llamacpp	llama.cpp	Local OpenAI-compatible preset; API key is optional
custom ID	Custom Compatible	Any OpenAI-compatible endpoint with your chosen provider ID

openai and openai-codex are intentionally separate routes. openai uses an OpenAI Platform API key. openai-codex uses ChatGPT/Codex OAuth state from the OpenAI connect flow.

API key setup

Manual API-key setup uses matching provider IDs in both files.

config.yaml:

version: 1

providers:
  - id: openai
  - id: anthropic

model:
  primary: openai/gpt-5

utility_model: openai/gpt-5-mini

auth.yaml:

openai:
  type: api
  access: "sk-..."

anthropic:
  type: api
  access: "sk-ant-..."

The key under auth.yaml must match the provider id from config.yaml.

Compatible providers

OpenAI-compatible providers use an id plus base_url in config.yaml.

version: 1

providers:
  - id: openrouter
    base_url: https://openrouter.ai/api/v1
  - id: ollama
    base_url: http://localhost:11434/v1

model:
  primary: openrouter/openai/gpt-5

auth.yaml:

openrouter:
  type: api
  access: "sk-or-..."

ollama:
  type: api
  access: ""

Local providers such as Ollama, LM Studio, and llama.cpp can use an empty API key when the local endpoint does not require authentication.

Model routes

After a provider is connected, select models from inside the TUI:

• /model changes the primary model route.
• /utility-model changes the background utility model route.
• /reviewer-model changes the model used for reviewer turns.

For saved defaults, set the same routes in config.yaml:

model:
  primary: anthropic/claude-sonnet-4-6

utility_model: openai/gpt-5-mini

workflow:
  review_model:
    primary: openai/gpt-5-mini

See Model Routing for routing behavior, model_overrides, reasoning controls, and model catalog refresh details.

Provider metadata

KodaCode refreshes remote model catalogs and caches them in the data directory. Some compatible providers return incomplete model metadata. When context windows, output caps, capabilities, or pricing are missing, add model_overrides in config.yaml.

model_overrides:
  - ref: qwencloud/qwen3.6-plus
    name: Qwen3.6 Plus
    context_size: 1000000
    max_input_tokens: 1000000
    max_output_tokens: 64000
    default_output_tokens: 16000
    reasoning: true
    tool_calls: true
    vision: true

See Configuration for the full persisted configuration surface.

Web search providers

Model providers are separate from web_search providers. Configure Exa or Parallel under web_search.providers, not top-level providers:.

See Web Search for web-search provider examples.