Skip to main content
Collabase Brain is the AI assistant built into the editor. It lets users summarise long pages, generate content from a prompt, translate text, improve writing, and more — without leaving Collabase. You configure and enable these features under Settings → AI.

​
Collabase Brain

Brain is the master switch for all AI features across the platform. When Brain is disabled, the AI command in the editor is hidden and no AI requests are sent to any provider. Toggle Collabase Brain on or off in Settings → AI. All other AI settings take effect only when Brain is enabled.

​
Supported AI providers

Collabase works with both local (self-hosted) and cloud providers. Choose the option that fits your privacy and infrastructure requirements.

Ollama

Run models locally on your server. No data leaves your infrastructure. Ideal for air-gapped or privacy-sensitive deployments.

OpenAI

Use GPT-4o, GPT-4o-mini, or any model available through the OpenAI API.

Google Gemini

Use Gemini 2.0 Flash and other Gemini models via the Google AI API.

Azure AI Foundry

Enterprise Azure OpenAI models deployed in your Azure subscription. Supports custom endpoint URLs.

Groq

Fast cloud inference. Groq offers a generous free tier and low-latency responses.
Select Disabled to turn off AI processing entirely without disabling Collabase Brain globally. This is useful when you want to keep Brain enabled in principle but pause the provider connection temporarily.

​
Configuring a provider

Navigate to Settings → AI and select your provider from the grid.

​
Ollama (local)

Ollama runs models on your own server. Install Ollama separately and pull the models you want to use before configuring it here.
FieldDescriptionDefault
Ollama Base URLThe URL where your Ollama server is running.http://localhost:11434
ModelThe name of the Ollama model to use.llama3.2
When using Ollama, all AI processing stays entirely on your server. No content is sent to any external service. This is the recommended option for organisations with strict data residency requirements.
After entering the Ollama Base URL, click Test Connection to verify that Collabase can reach your Ollama server and that the specified model is available.

​
OpenAI

FieldDescriptionExample
API KeyYour OpenAI API key. Stored encrypted, never displayed after saving.sk-...
ModelThe OpenAI model name to use.gpt-4o-mini

​
Google Gemini

FieldDescriptionExample
API KeyYour Google AI Studio API key.AIza...
ModelThe Gemini model name to use.gemini-2.0-flash

​
Azure AI Foundry

FieldDescriptionExample
Azure Endpoint URLThe base URL of your Azure AI Foundry deployment.https://resource.services.ai.azure.com/openai
API KeyYour Azure API key....
ModelThe deployed model name as configured in Azure.gpt-4o

​
Groq

FieldDescriptionExample
API KeyYour Groq API key from console.groq.com.gsk_...
ModelThe Groq model name.llama-3.3-70b-versatile

​
Saving and testing

After filling in your provider’s fields, click Test Connection to verify that the credentials and model configuration are correct. Then click Save to apply the settings.
API keys are encrypted before being stored in the database. After saving, the key is masked in the UI. To rotate a key, enter the new value and save again.

​
The AI command in the editor

Once Brain is enabled and a provider is configured, users can trigger AI actions anywhere in the editor using the /ai slash command. Type /ai on any line to open the AI action menu. Available actions include:
  • Summarise — Compress the current page or selection into a concise summary.
  • Generate — Write content from a prompt.
  • Translate — Translate selected text to a target language.
  • Improve writing — Refine grammar, clarity, and tone.
  • Fix spelling & grammar — Correct errors in the selection.
  • Make shorter / Make longer — Adjust the length of the selected text.
  • Continue writing — Extend the text from the current cursor position.
AI actions operate on the selected text when text is selected, or on the paragraph at the cursor when nothing is selected.
Collabase supports Retrieval-Augmented Generation (RAG) to ground AI responses in your actual content. When RAG is enabled, page content is chunked into 512-token segments and indexed as vector embeddings. The Brain chat feature uses these embeddings to retrieve relevant context before generating a response. Enable RAG / Semantic Search in Settings → AI and choose an embedding source:
Collabase reuses the same provider and model configured above for generating embeddings. This is the simplest option and works well when your AI provider supports embedding models.
Embeddings are generated asynchronously when pages are saved. Newly created or updated pages become available for semantic search after the indexing job completes. Indexing typically takes a few seconds per page.