Provider Catalog

Every provider ZeroClaw ships with. For each: what it talks to, config shape, and notes.

See Configuration for universal fields.

Native

Anthropic

[providers.models.claude]
kind = "anthropic"
model = "claude-haiku-4-5-20251001"     # or claude-sonnet-4-6, claude-opus-4-7
api_key = "sk-ant-..."                   # or "sk-ant-oat-..." for OAuth

Supports OAuth tokens (sk-ant-oat*) from Claude Pro/Team subscriptions — no separate API billing. Streaming, tool calls, vision, and reasoning all supported.

OpenAI

[providers.models.gpt]
kind = "openai"
model = "gpt-4o-mini"
api_key = "sk-..."

GPT-4o, GPT-5, o-series reasoning models. Reasoning tokens surfaced as ReasoningDelta events; see Streaming.

Ollama

[providers.models.local]
kind = "ollama"
base_url = "http://localhost:11434"
model = "qwen3.6:35b-a3b"
think = false                     # disable chain-of-thought on reasoning models

Local inference. Uses Ollama's native /api/chat. Schema-based structured output via format parameter (reliability varies by model). No API key needed.

Bedrock

[providers.models.bedrock]
kind = "bedrock"
region = "us-east-1"
model = "anthropic.claude-3-5-sonnet-20241022-v2:0"
# uses AWS credential chain — IAM role, env vars, ~/.aws/credentials

AWS-hosted Claude, Llama, Titan, and others. Auth via the standard AWS credentials chain — no explicit key in config needed if your environment is set up for AWS.

Gemini

[providers.models.gemini]
kind = "gemini"
model = "gemini-2.5-pro"
api_key = "..."

Google's Gemini API. Supports vision and pre-executed grounded search (see Streaming for PreExecutedToolCall events).

Gemini CLI

[providers.models.gemini-cli]
kind = "gemini-cli"
model = "gemini-2.5-pro"

Shells out to the gemini CLI. No API key in config — uses whatever auth the CLI has.

Azure OpenAI

[providers.models.azure]
kind = "azure-openai"
base_url = "https://my-resource.openai.azure.com"
deployment = "gpt-4o"
api_version = "2024-10-01-preview"
api_key = "..."

Copilot

[providers.models.copilot]
kind = "copilot"
model = "gpt-4o"
# authenticate once via `zeroclaw onboard` — the wizard handles the OAuth flow
# and stores a token at ~/.config/zeroclaw/copilot.json

Uses a GitHub Copilot subscription for agent inference. OAuth-managed.

Claude Code

[providers.models.cc]
kind = "claude-code"

Delegates turns to a Claude Code session over MCP. Useful for code-heavy workflows; inherits Claude Code's tool allow-lists and project context.

Telnyx

[providers.models.voice-brain]
kind = "telnyx"
model = "..."
api_key = "..."

Voice-oriented AI endpoint. Pair with the clawdtalk channel for real-time SIP calls.

KiloCLI

[providers.models.kilo]
kind = "kilocli"
model = "..."

Local inference via KiloCLI.

OpenAI-compatible (compatible.rs, ~20+ endpoints)

One Rust impl reused for every endpoint that speaks OpenAI chat completions. Pattern:

[providers.models.<name>]
kind = "openai-compatible"
base_url = "<endpoint>"
model = "<model-id>"
api_key = "<key>"

Verified endpoints:

Any endpoint that claims OpenAI chat-completions compatibility should work — if it doesn't, file an issue with a minimal reproducer.

Meta

OpenRouter

[providers.models.openrouter]
kind = "openrouter"
model = "anthropic/claude-sonnet-4-20250514"   # openrouter's vendor/model form
api_key = "sk-or-..."

Routes through OpenRouter's fan-out layer. Use when you want one billing relationship across many models.

Reliable (fallback chain)

[providers.models.main]
kind = "reliable"
fallback_providers = ["claude", "openrouter", "local"]

See Fallback & routing.

Router (task-hint)

[providers.models.brain]
kind = "router"
default = "haiku"
routes = [
    { hint = "reasoning", provider = "deepseek-r1" },
    { hint = "vision",    provider = "gemini" },
]

See Fallback & routing.

Something missing?

If the endpoint you want isn't listed, it's probably OpenAI-compatible — try kind = "openai-compatible" with the appropriate base_url. If it's not OpenAI-compatible and needs its own implementation, see Custom providers.