Files
ironclaw/docs/drafts/providers/index.mdx
2026-04-09 14:18:30 +02:00

160 lines
4.2 KiB
Plaintext

---
title: LLM Providers
sidebarTitle: Overview
description: Choose your AI model provider
---
IronClaw supports multiple LLM backends. Choose the provider that best fits your needs.
## Provider Comparison
| Provider | Backend | Local | Privacy | Cost | Best For |
|----------|---------|-------|---------|------|----------|
| **NEAR AI** | `nearai` | ✗ | ★★★ | $ | Default, easy setup |
| **Anthropic** | `anthropic` | ✗ | ★★★ | $$ | Claude models |
| **OpenAI** | `openai` | ✗ | ★★ | $$ | GPT models |
| **Ollama** | `ollama` | ✓ | ★★★★★ | Free | Local inference |
| **Tinfoil** | `tinfoil` | ✗ | ★★★★★ | $$ | TEE privacy |
| **OpenRouter** | `openai_compatible` | ✗ | ★★ | $ | Model variety |
| **Moonshot** | `openai_compatible` | ✗ | ★★★ | $$ | Kimi K2.5 |
## Quick Selection Guide
<Tabs>
<Tab title="Easiest">
**NEAR AI** — Default provider, works out of the box
- Browser OAuth or API key
- Multiple models
- No credit card required to start
</Tab>
<Tab title="Most Private">
**Ollama** — Run models locally
- No data leaves your machine
- Free
- Requires GPU for larger models
</Tab>
<Tab title="Best Models">
**Anthropic Claude** — Industry-leading reasoning
- Excellent for complex tasks
- Long context window
- Higher cost
</Tab>
<Tab title="Model Variety">
**OpenRouter** — Access 300+ models
- Single API key
- Mix of commercial and open models
- Pay-as-you-go
</Tab>
</Tabs>
## Available Providers
<CardGroup cols={2}>
<Card title="NEAR AI" icon="star" href="/providers/nearai">
Default provider. Browser OAuth or API key. Multiple models including Claude.
</Card>
<Card title="Anthropic" icon="triangle" href="/providers/anthropic">
Claude models directly. Best reasoning and long context.
</Card>
<Card title="OpenAI" icon="circle" href="/providers/openai">
GPT-4o, GPT-4o-mini, o3-mini. Direct API access.
</Card>
<Card title="Ollama" icon="download" href="/providers/ollama">
Local inference. Free, private. Llama, Mistral, Qwen, and more.
</Card>
<Card title="OpenAI-Compatible" icon="layers" href="/providers/openai-compatible">
OpenRouter, Together AI, Fireworks, vLLM, LiteLLM, LM Studio.
</Card>
<Card title="Tinfoil" icon="shield" href="/providers/tinfoil">
Hardware-attested TEE. Neither Tinfoil nor cloud can see prompts.
</Card>
<Card title="Moonshot" icon="sparkles" href="/providers/moonshot">
Kimi K2.5 with 256K context. Advanced reasoning and long-context models.
</Card>
</CardGroup>
## Switching Providers
Change providers by updating environment variables and restarting:
```bash
# Edit ~/.ironclaw/.env
export LLM_BACKEND=anthropic
export ANTHROPIC_API_KEY=sk-ant-...
# Restart IronClaw
ironclaw run
```
Or re-run the wizard:
```bash
ironclaw onboard --skip-auth
```
## Configuration
All providers are configured via environment variables:
```bash
# In ~/.ironclaw/.env
LLM_BACKEND=anthropic
ANTHROPIC_API_KEY=sk-ant-...
ANTHROPIC_BASE_URL=https://api.anthropic.com # optional
```
See [Configuration](/setup/configuration) for the complete reference.
## Privacy Considerations
| Provider | Data Handling |
|----------|---------------|
| **NEAR AI** | Prompts/responses sent to NEAR AI |
| **Anthropic** | Prompts/responses sent to Anthropic |
| **OpenAI** | Prompts/responses sent to OpenAI |
| **Ollama** | No external data transmission |
| **Tinfoil** | Encrypted in TEE, provider cannot see |
For maximum privacy, use **Ollama** (local) or **Tinfoil** (TEE).
## Cost Estimation
Rough cost per 1K tokens (input + output):
| Provider | Cost |
|----------|------|
| Ollama | Free (GPU/electricity only) |
| OpenAI GPT-4o-mini | ~$0.60 |
| NEAR AI | ~$1-3 |
| OpenAI GPT-4o | ~$5-10 |
| Anthropic Claude | ~$3-15 |
Actual costs vary by model and usage.
## Next Steps
Choose your provider:
<CardGroup cols={3}>
<Card title="Start with NEAR AI" icon="rocket" href="/providers/nearai">
Default provider — easiest to set up
</Card>
<Card title="Go Local" icon="lock" href="/providers/ollama">
Maximum privacy with Ollama
</Card>
<Card title="Use Claude" icon="brain" href="/providers/anthropic">
Best reasoning capabilities
</Card>
</CardGroup>