Model Configuration

SailFish supports multiple AI model configurations simultaneously. You can switch between them on the fly depending on the task. This page covers all model management features.

Supported Providers

SailFish includes preset templates for these providers — click to auto-fill the API URL and default model:

Also supports Doubao, Zhipu, Kimi, Grok, Mistral, and any other provider with an OpenAI-compatible API.

No API Key yet? DeepSeek offers free credits on sign-up, has strong models, and is the best value to get started.

Adding a Model Configuration

1. Open Settings → AI Model Configuration
2. Click “Add”
3. Select a provider template (auto-fills the API URL) or enter details manually
4. Fill in the fields:

5. Click Save
6. Click “Use” to make this the active model

Multi-Model Strategy

We recommend configuring 2–3 models and switching by scenario:

Switch models anytime using the selector at the top of the chat area — the current conversation context is preserved.

Hybrid Multimodal (Linked Vision Model)

Many powerful reasoning models (like DeepSeek R1) don’t support image input, while vision models may not match them at complex reasoning. SailFish’s “linked model” feature gives you the best of both worlds.

How It Works

1. You send a message containing an image or screenshot
2. SailFish detects the image in the message
3. Checks whether the current primary model supports vision
4. If not → automatically routes this request to the linked vision model
5. After the vision model responds, subsequent text-only messages go back to the primary model

Setup Steps

1. Open Settings → AI Model Configuration
2. Make sure you have a vision-capable model (e.g. GPT-4o) with “Type” set to Vision
3. Edit your primary text model (e.g. DeepSeek R1), set “Type” to General
4. In the “Linked Vision Model” dropdown, select the vision model
5. Ensure “Auto Vision Model Switch” is enabled (Settings → General)

Typical Usage

Primary model: DeepSeek R1. Linked vision model: GPT-4o.

Tips

• Pick the strongest reasoner as primary (DeepSeek R1, o1), and the best vision model as linked (GPT-4o, Claude Sonnet)
• If your primary model already supports vision (e.g. GPT-4o), no linked model is needed
• Switching happens per-request and doesn’t break conversation flow

Proxy Settings

If you need a proxy to reach overseas AI services (OpenAI, Claude, Gemini, etc.), you can configure one per model:

1. Edit the model configuration
2. Enter the proxy address in the “Proxy” field:
   • HTTP: http://127.0.0.1:7890
   • SOCKS5: socks5://127.0.0.1:1080
3. Save

Proxy settings are per-configuration — DeepSeek can go direct while OpenAI uses a proxy, with no interference.

Ollama (Local Deployment)

Run AI models locally for free, with no internet required:

Install Ollama

1. Visit ollama.com to download and install
2. Pull a model in your terminal:

ollama pull qwen2.5:7b    # recommended for Chinese
ollama pull llama3.1:8b    # strong for English

Configure in SailFish

1. Add a new AI configuration, select the Ollama template
2. API URL auto-fills to http://localhost:11434
3. Leave API Key blank
4. Enter the model name you pulled (e.g. qwen2.5:7b)

Note: Local model capability depends on model size and your hardware. 7B models work well for everyday tasks; for complex operations, pair with a cloud model.

Common Issues

API Key entered but getting 401 errors

• Check the key is complete (no extra spaces or missing characters)
• Confirm the account has credit/balance remaining
• Some providers require a deposit before keys work

What model name should I enter?

• Use the model ID from the provider’s documentation (case-sensitive)
• e.g. gpt-4o (not GPT-4o), deepseek-chat (not DeepSeek Chat)

AI behavior changed after switching models

• Different models have very different capabilities — this is expected
• Smaller models may struggle with complex multi-step tasks
• If a task doesn’t work well with the current model, try a more capable one