Skip to content

LLM Providers

DAIV supports the following LLM providers:

You can mix providers — for example, use OpenRouter for the main agent and a direct provider for a specific model override.

How models are specified

Models use a prefix system to identify the provider:

Prefix Provider Example
openrouter: OpenRouter openrouter:anthropic/claude-sonnet-4.6
claude Anthropic (direct) claude-sonnet-4.6
gpt-4, gpt-5, o4 OpenAI (direct) gpt-5.3-codex
gemini Google Gemini (direct) gemini-2.5-pro-preview-05-06

DAIV resolves the provider automatically from the model name. No extra configuration is needed beyond setting the API key for the provider you want to use.

Tip

In the dashboard, the agent picker offers a live model dropdown built from each enabled provider's API (cached for 15 minutes), so you can choose a model without memorizing its full spec. Free-text entry remains available as a fallback.

Default models

Role Model Provider
Main agent Claude Sonnet 4.6 OpenRouter
Max mode Claude Opus 4.6 OpenRouter
Explore subagent Claude Haiku 4.5 OpenRouter
Fallback GPT 5.3 Codex OpenRouter

All defaults route through OpenRouter, so only OPENROUTER_API_KEY is required for a basic setup. Override models per repository in .daiv.yml — see Repository Config.

OpenRouter

OpenRouter is the default provider for DAIV. It provides access to models from multiple vendors with built-in fallback support.

Setup:

  1. Obtain an API key from OpenRouter Settings
  2. Set the environment variable:
    Text Only
    1
    OPENROUTER_API_KEY=your-api-key-here

Model format:

Use the openrouter: prefix followed by the model name from OpenRouter:

Text Only
1
2
openrouter:anthropic/claude-sonnet-4.6
openrouter:openai/gpt-5.3-codex

Per-repository override (.daiv.yml):

YAML
1
2
3
models:
  agent:
    model: "openrouter:anthropic/claude-sonnet-4.6"

OpenAI

Setup:

  1. Obtain an API key from OpenAI
  2. Set the environment variable:
    Text Only
    1
    OPENAI_API_KEY=your-api-key-here

Model format:

Use the model name from OpenAI directly (no prefix needed):

Text Only
1
2
gpt-5.3-codex
o4-mini

Per-repository override (.daiv.yml):

YAML
1
2
3
models:
  agent:
    model: "gpt-5.3-codex"

Anthropic

Setup:

  1. Obtain an API key from Anthropic
  2. Set the environment variable:
    Text Only
    1
    ANTHROPIC_API_KEY=your-api-key-here

Model format:

Use the model name from Anthropic directly (no prefix needed):

Text Only
1
2
claude-sonnet-4.6
claude-opus-4.6

Per-repository override (.daiv.yml):

YAML
1
2
3
models:
  agent:
    model: "claude-sonnet-4.6"

Warning

We love Anthropic, but unfortunately their API is very unstable and often returns errors. Also, the rate limits could be exceeded very quickly.

Google Gemini

Setup:

  1. Obtain an API key from AI Studio
  2. Set the environment variable:
    Text Only
    1
    GOOGLE_API_KEY=your-api-key-here

Model format:

Use the model name from Gemini directly (no prefix needed):

Text Only
1
2
gemini-2.5-pro-preview-05-06
gemini-2.4-flash-preview-04-17

Per-repository override (.daiv.yml):

YAML
1
2
3
models:
  agent:
    model: "gemini-2.5-pro-preview-05-06"

Custom providers

Beyond the four providers above, admins can register additional providers from the Configuration UI at /dashboard/configuration/ (under the providers section). This is mainly for self-hosted, OpenAI-compatible endpoints — vLLM, llama.cpp, LiteLLM, or an internal gateway — but applies to any of the supported wire protocols.

When you add a provider, you choose:

Field Purpose
Slug A short identifier that becomes the model prefix (your-slug:model_name). Must start with a lowercase letter; lowercase letters, digits, -, _; max 32 chars. Immutable after creation.
Provider type The wire protocol used to talk to the server — one of OpenAI, Anthropic, Google Gemini, or OpenRouter.
Base URL The endpoint to call instead of the provider's default (for example, your vLLM server). For OpenAI-typed servers, include the version segment (e.g. /v1).
API key The credential for the endpoint (encrypted at rest).
Extra headers A JSON object of additional request headers (e.g. {"X-Foo": "bar"}). Agent-managed headers take precedence.
Use Responses API Only honored for OpenAI-typed providers. Enable for servers that expose /v1/responses; leave off for /v1/chat/completions-only servers (vLLM, llama.cpp, most LiteLLM setups).
Verify TLS certificates Disable only for self-hosted endpoints behind an internal or self-signed CA. Skipping verification exposes the connection to MITM attacks — prefer mounting the CA into the DAIV containers when possible. Honored only for OpenAI- and OpenRouter-typed providers.

Once the row exists, reference its models using the slug as a prefix:

Text Only
1
your-slug:model_name

You can use this prefix anywhere a model is specified, including the .daiv.yml model overrides.

Note

The four built-in providers (OpenRouter, OpenAI, Anthropic, Google Gemini) are locked: their slug and provider type cannot be changed and the rows cannot be deleted. Custom providers you add are fully editable and removable.

Admin only

Managing providers requires an admin account. See Automation: LLM Providers for how the built-in providers map to environment variables.