CANChat chatbot
Developed by SSC, CANChat is a multilingual conversational chatbot designed to enhance productivity. It serves as an in-house alternative to commercial generative AI tools. Government of Canada (GC) workers can use it for numerous purposes, such as:
- drafting documents
- editing content
- assisting with research
- many more functions
Status
CANChat is in a pilot phase with selected departments and agencies.
For more information, visit CANChat GCxchange (available only on the GC network).
Pilot participants must read and accept the CANChat Terms of Use to access the tool. It provides essential information on permitted and prohibited activities, ensuring a clear understanding of usage requirements.
User feedback will help us to refine CANChat and build a future enterprise version.
Protecting user data and privacy
In alignment with GC policies and standards, CANChat incorporates various guardrails and safeguards to protect user data. The large language models (LLM) in CANChat do not use prompts or data for training, which protects user privacy.
Providing generative AI models for a variety of tasks
CANChat provides access to a wide range of generative AI models. This includes the GC’s own LLMs, as well as third-party options.
Unlike commercial offerings, CANChat enables users to switch among the models. Users can select the most suitable model for their work, considering data residency and sovereignty requirements.
| Model source | Description | Suitable tasks |
|---|---|---|
| GC | A fine-tuned LLM (based on Meta Llama) built specifically for the GC, fine-tuned with Canadian content for greater contextual alignment | Well-suited for drafting professional correspondence, developing written content and generating clear, well-structured answers to queries, especially those requiring a Canadian context |
| Cohere | A made-in-Canada LLM, trained on diverse, high-quality datasets that reflect Canada’s unique cultural, linguistic and regulatory landscape | General purpose model for text generation, information retrieval and tool calling (e.g., agentic AI), supporting both English and French, and ensuring compliance with Canada’s Official Languages Act |
| Google (Gemini) |
A general purpose, cost-efficient AI model that blends high performance with:
|
Best for fast, cost-efficient AI tasks that need scale and versatility, such as chatbots, summarization, multimodal queries (text, images) and code help; ideal when low latency and large context handling are more important than maximum analytical power |
| OpenAI (GPT 4o) | Similar to GPT-4, but optimized for faster processing | Suitable for tasks that require the depth of Chat GPT-4, but where faster response times are crucial, such as time-sensitive tasks or high-frequency requests; also for tasks that involve interpreting or generating images |
| OpenAI (GPT 5) | The latest version of OpenAI’s model, improving on previous versions |
Good at reading, writing and reasoning across many types of tasks:
|
| Meta (Llama) | An AI model that processes and understands both text and images, enabling advanced multimodal capabilities such as image analysis, caption generation and text-based reasoning | Ideal for analyzing images, generating descriptions, extracting insights from visual content, and combining text and image processing |
Building a trusted, sovereign AI ecosystem
CANChat is a step forward in building a trusted, sovereign AI ecosystem that reflects Canadian values and protects the data entrusted to the GC.