-
-
Notifications
You must be signed in to change notification settings - Fork 0
LLM Providers
Open-Write supports 22 LLM providers out of the box, including cloud APIs and local inference engines.
| Provider | Base URL | Notes |
|---|---|---|
| OpenRouter | https://openrouter.ai/api/v1 |
Many models behind one key |
| Provider | Base URL | Notes |
|---|---|---|
| OpenAI | https://api.openai.com/v1 |
GPT-4o, GPT-4.1, o3-mini, etc. |
| Anthropic | https://api.anthropic.com/v1 |
Claude Sonnet 4, Claude 3.5 Haiku, Opus |
| Google AI | https://generativelanguage.googleapis.com/v1beta/openai |
Gemini 2.5 Pro/Flash, Gemma 3 |
| Mistral | https://api.mistral.ai/v1 |
Mistral Large, Small, Pixtral, Codestral |
| Groq | https://api.groq.com/openai/v1 |
Ultra-fast inference, Llama 3.3 70B |
| xAI | https://api.x.ai/v1 |
Grok 3, Grok 2 |
| Together AI | https://api.together.xyz/v1 |
Llama 3.3/3.1, DeepSeek V3, Qwen 2.5 |
| Fireworks AI | https://api.fireworks.ai/inference/v1 |
Llama, DeepSeek, Qwen |
| DeepInfra | https://api.deepinfra.com/v1/openai |
Llama, DeepSeek, Qwen |
| Perplexity | https://api.perplexity.ai |
Sonar Pro, Sonar, Sonar Reasoning |
| Provider | Base URL | Notes |
|---|---|---|
| DeepSeek | https://api.deepseek.com/v1 |
DeepSeek V3, R1 |
| GLM (Zhipu) | https://open.bigmodel.cn/api/paas/v4 |
GLM-4 Plus, Flash, Long |
| Qwen (Alibaba) | https://dashscope.aliyuncs.com/compatible-mode/v1 |
Qwen Max, Plus, Turbo |
| Moonshot (Kimi) | https://api.moonshot.cn/v1 |
128K, 32K, 8K context |
| MiniMax | https://api.minimax.chat/v1 |
abab6.5, abab5.5 |
| Baichuan | https://api.baichuan-ai.com/v1 |
Baichuan 4, 3-Turbo |
| StepFun | https://api.stepfun.com/v1 |
Step-2, Step-1 |
| SiliconFlow | https://api.siliconflow.cn/v1 |
DeepSeek, Qwen, Llama |
| MiMo | (user-configured) | User sets base URL |
| Provider | Default Endpoint | Notes |
|---|---|---|
| LM Studio | http://localhost:1234/v1 |
Auto-detects loaded models |
| Ollama | http://localhost:11434/v1 |
Auto-detects installed models |
Open-Write ships with a Recommended Models panel that works even before you configure any provider keys. Models are tagged by:
-
Tier:
free/budget/standard/premium -
Strengths:
prose,reasoning,fast,bilingual,long-context,open-source,unfiltered, etc.
The catalog includes 26+ curated models across all providers, plus 12 local models for LM Studio and Ollama.
- Open Settings (gear icon)
- Scroll to LLM Providers
- Find your provider, paste your API key
- Click Test to verify the connection
- Select a model from the Recommended Models panel or type one manually
LM Studio and Ollama let you run models on your own hardware with no API key:
- Install LM Studio or Ollama
- Download a model (e.g.,
ollama pull llama3.3:70b) - Open-Write auto-detects running models at the default endpoint
- All catalog models for local providers are tagged as
freetier
| Model | VRAM | Strengths |
|---|---|---|
| Llama 3.3 70B | 24+ GB | Prose, reasoning |
| Qwen 2.5 72B | 48+ GB | Prose, bilingual, long-context |
| Mistral Nemo 12B | 8+ GB | Fast, compact, good prose |
| Phi-4 14B | 8+ GB | Fast, reasoning |
| DeepSeek R1 70B | 48+ GB | Reasoning, plot planning |
| Gemma 2 9B | 6+ GB | Fast, modest hardware |
The pipeline can use different models for different phases:
- Writer model — used for architect, writer, bible, voice phases
- Critic model — used for the 5 critics + editorial evaluation
- Planner model — used by the harness layer planner
Configure these in Settings under Pipeline Model Routing. Each defaults to the primary model if unset.
Any OpenAI-compatible endpoint works. Add a custom provider in Settings with:
- A base URL (must end in
/v1or equivalent) - An API key (if required)
- A model name (or comma-separated list)
Welcome to the Open-Write Studio wiki!
This wiki covers everything you need to know about Open-Write — from installation to architecture to contributing.
New here? Start with Getting Started.