> ## Documentation Index
> Fetch the complete documentation index at: https://docs.openpawz.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Providers

> 10 AI providers and any OpenAI-compatible API — local or cloud, cheap or powerful.

# Providers

Providers are the AI backends that power your agents. OpenPawz supports **10 providers** and any OpenAI-compatible API.

<Note>
  Mix local models (Ollama) with cloud providers. OpenPawz auto-falls back to the next provider if one goes down.
</Note>

***

## Supported providers

<Columns cols={3}>
  <Card title="Ollama" href="/providers/ollama" icon="server">
    Local inference — no API key needed
  </Card>

  <Card title="OpenAI" href="/providers/openai" icon="bolt">
    GPT-4o, o1, o3, o4-mini
  </Card>

  <Card title="Anthropic" href="/providers/anthropic" icon="brain">
    Claude 4 Opus, Sonnet, Haiku
  </Card>

  <Card title="Google" href="/providers/google" icon="google">
    Gemini 2.5 Pro, Flash
  </Card>

  <Card title="OpenRouter" href="/providers/openrouter" icon="route">
    200+ models via unified API
  </Card>

  <Card title="DeepSeek" href="/providers/deepseek" icon="magnifying-glass">
    DeepSeek V3, R1, Coder
  </Card>

  <Card title="Grok" href="/providers/grok" icon="xmark">
    xAI's Grok models
  </Card>

  <Card title="Mistral" href="/providers/mistral" icon="wind">
    Mistral Large, Codestral
  </Card>

  <Card title="Moonshot" href="/providers/moonshot" icon="moon">
    Kimi and Moonshot models
  </Card>

  <Card title="Custom" href="/providers/custom" icon="plug">
    Any OpenAI-compatible API
  </Card>
</Columns>

## Adding a provider

1. Go to **Settings → Providers**
2. Click **Add Provider**
3. Select the provider type
4. Enter your API key
5. Optionally set a default model and custom base URL

## Model routing

OpenPawz can automatically route requests to different models based on the context:

| Setting              | Purpose                                                         |
| -------------------- | --------------------------------------------------------------- |
| **Boss model**       | Expensive, powerful — used for orchestrator bosses              |
| **Worker model**     | Cheap, fast — used for channel responses and workers            |
| **Specialty models** | Per-specialty overrides (e.g., `coder` → `gemini-2.5-pro`)      |
| **Agent models**     | Per-agent overrides (highest priority)                          |
| **Cheap model**      | Simple tasks auto-select this model                             |
| **Auto-tier**        | Automatically pick cheap vs. expensive based on task complexity |

**Resolution priority:** agent\_models > specialty\_models > role-based (boss/worker) > fallback

## Smart prefix matching

When you type a model name, OpenPawz auto-routes to the right provider:

| Pattern                                     | Routes to                           |
| ------------------------------------------- | ----------------------------------- |
| `claude*` / `anthropic*`                    | Anthropic                           |
| `gemini*` / `google*`                       | Google                              |
| `gpt*` / `o1*` / `o3*` / `o4*`              | OpenAI                              |
| Contains `/` (e.g., `meta-llama/llama-3.1`) | OpenRouter                          |
| Contains `:` (e.g., `llama3.1:8b`)          | Ollama                              |
| Other                                       | Default provider → first configured |

## Provider fallback

If a provider returns a billing, authentication, or rate limit error, OpenPawz automatically tries other configured providers before failing. This ensures your agents stay responsive even if one provider has issues.

## Budget enforcement

Set a daily budget in **Settings → Engine**:

| Threshold | Action               |
| --------- | -------------------- |
| 50%       | Warning notification |
| 75%       | Warning notification |
| 90%       | Warning notification |
| 100%      | Requests blocked     |

Set to `0` to disable budget limits.

## Multiple providers

You can configure as many providers as you want. Benefits:

* Use different models for different agents
* Automatic fallback if one provider is down
* Mix local (Ollama) and cloud providers
* Use cheap models for routine tasks, expensive for complex ones

## Provider security

Every provider connection is protected by multiple security layers:

| Layer                  | Protection                                                                                                                |
| ---------------------- | ------------------------------------------------------------------------------------------------------------------------- |
| **TLS pinning**        | Custom `rustls` config with Mozilla root CAs only — OS trust store excluded to prevent MITM from compromised CAs          |
| **Request signing**    | Every outbound request SHA-256 signed (`provider ‖ model ‖ timestamp ‖ body`) and logged to a 500-entry audit ring buffer |
| **Key zeroing**        | API keys wrapped in `Zeroizing<String>` — automatically overwritten with zeros in RAM on provider drop                    |
| **Credential vault**   | Keys encrypted with XOR (32-byte key in OS keychain) at rest, decrypted only at call time                                 |
| **Circuit breaker**    | 5 consecutive failures → 60-second cooldown window to prevent cascading failures                                          |
| **Retry with backoff** | Exponential backoff with ±25% jitter (1s → 30s), respects `Retry-After` headers                                           |

→ Deep dive: [Security reference](/reference/security)