AI Models & Providers

dassi supports multiple AI models. You can use managed credits (default, no setup required), bring your own API key for a named provider, or connect any OpenAI-compatible endpoint.

Managed credits (default)

When you sign in, dassi automatically provisions managed credits. No API key is needed — just pick a model and start working.

Available models depend on your plan:

Model	Free Trial	Paid plans
Gemini 3.5 Flash	yes	yes
Gemini 3 Pro	yes	yes
Claude Haiku 4.5	—	yes
Claude Sonnet 4.5	—	yes
Claude Opus 4.6	—	yes
Gemini 3 Flash	—	yes
DeepSeek V3.2	—	yes
GLM 4.7	—	yes

The default model is Gemini 3.5 Flash. You can switch models in Settings at any time.

Choosing a model

• Gemini 3.5 Flash — fast and cost-efficient. Good all-around default choice
• Gemini 3 Pro — strong general-purpose model, good for complex reasoning
• Claude Sonnet 4.5 — excellent for writing, analysis, and complex reasoning
• Claude Haiku 4.5 — fastest response times, good for simple tasks
• Claude Opus 4.6 — most capable, best for complex multi-step tasks
• Gemini 3 Flash — very fast, good for quick questions

Bring your own key (BYOK)

If you prefer to use your own OpenAI API key:

1. Open Settings (gear icon in the sidebar toolbar)
2. Select BYOK as the billing mode
3. Enter your OpenAI API key
4. Choose a model:

Model	Description
GPT-5	Most capable OpenAI model
GPT-5 Mini	Balanced performance and cost
GPT-5 Nano	Fastest, most affordable

Getting an OpenAI API key

1. Sign up at platform.openai.com
2. Navigate to API Keys in your account settings
3. Click Create new secret key
4. Copy the key and paste it in dassi’s settings

Custom (OpenAI-compatible) endpoint

You can connect any server that speaks the OpenAI Chat Completions API, including:

• Azure OpenAI — a model deployed in your Azure subscription
• Ollama — local models running on your machine
• LiteLLM — a proxy that routes to many providers under one endpoint
• vLLM and other self-hosted inference servers

Connecting

1. Open Settings (gear icon in the sidebar toolbar) and select the Connection tab
2. Find the Custom (OpenAI-compatible) card and click Add endpoint
3. Enter the Base URL of your endpoint — for example, https://my-resource.openai.azure.com/openai/v1 for Azure or http://localhost:11434 for a local Ollama server
4. Enter the Model ID — the exact identifier your server expects, e.g. gpt-4o, llama3.2, or mistral-7b
5. Enter your API key. For Ollama running locally, enter any non-empty string (e.g. ollama)
6. Click Connect — Chrome will ask to allow access to that endpoint, then dassi verifies the connection

Supported model families

Dassi accepts model IDs from these families: GPT, Claude, Gemini/Gemma, Grok, Kimi/Moonshot, DeepSeek, Llama, Qwen, Mistral/Mixtral, Phi, Command-R, GLM, Nemotron, and Yi. If your model ID doesn’t match a supported family, the connection will be blocked with a message telling you which families are accepted.

Notes

• Local servers using plain HTTP (e.g. http://localhost:11434) are allowed. Remote endpoints must use HTTPS.
• Dassi normalizes your base URL to end with /v1, so entering http://localhost:11434 or http://localhost:11434/v1 connects to the same endpoint.

How dassi uses models

dassi uses your selected model as the main reasoning engine. For certain visual tasks (like clicking at specific coordinates on a page), dassi automatically uses a faster secondary model to keep things quick and cost-efficient. You don’t need to configure this — it happens automatically.

Switching models

You can switch models at any time from the Settings page. The change takes effect for your next message. Your conversation history is preserved when switching models.