Skip to content

LLM Providers

Jump to bottom
Nick edited this page Jul 28, 2026 · 1 revision

LLM Providers

Open-Write supports 22 LLM providers out of the box, including cloud APIs and local inference engines.

Provider List

Aggregators

Provider Base URL Notes
OpenRouter https://openrouter.ai/api/v1 Many models behind one key

US Providers

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

Chinese Providers

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

Local Providers (No API Key)

Provider Default Endpoint Notes
LM Studio http://localhost:1234/v1 Auto-detects loaded models
Ollama http://localhost:11434/v1 Auto-detects installed models

Curated Model Catalog

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.

Configuring a Provider

  1. Open Settings (gear icon)
  2. Scroll to LLM Providers
  3. Find your provider, paste your API key
  4. Click Test to verify the connection
  5. Select a model from the Recommended Models panel or type one manually

Local Inference

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 free tier

Popular Local Models

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

Per-Phase Model Routing

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.

Custom Endpoints

Any OpenAI-compatible endpoint works. Add a custom provider in Settings with:

  • A base URL (must end in /v1 or 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.

Navigation

Clone this wiki locally