Getting Started with ALTAI
ALTAI is a platform for building synthetic datasets from your documents and fine-tuning models on that data.
What you can build
- Synthetic datasets — Turn your uploaded files (PDF, DOCX, RTF, HTML, TXT) into conversational training data using the Afterimage generation stack.
- Fine-tuned models — Train models (like Qwen3, Gemma, or Llama variants) on your datasets using Cinderflow, our purpose-built workflow runner.
- Other modes — Additional run modes like RAG and Zero-tune are available, with Agents and Evaluation coming soon.
Create your account
- Go to app.altai.dev and click Sign in to access the auth page.
- On the Sign Up tab, enter a valid email and a password (at least 8 characters), or sign in with Google or GitHub.
- After signing in, you will land on the Dashboard, where you can view your projects, runs, and credit balance.
Organizations and Teammates
ALTAI uses organization-based billing and collaboration. When you sign up, a "Personal Team" organization is automatically created for you.
- Roles: There are three types of roles within an organization: Owner, Admin, and Member.
- As the creator, you automatically become the Owner. You can invite teammates as Admins or Members, and Admins can also manage Members within the organization.
- You can create additional organizations to keep different teams or billing separate.
Projects and Workspace
Projects are bound to an organization. Each project has its own workspace consisting of:
- Runs — synthetic dataset jobs, fine-tuning jobs, etc.
- Files — documents you upload for generation and training.
Credits and Billing
Credits are Altai’s internal unit for consumption and are tied to your organization. When a project consumes credits, they are deducted from its parent organization's balance.
- Synthetic dataset generation consumes credits based on LLM token usage during generation.
- Fine-tuning consumes credits based on GPU job cost from the training system.
You can view your organization's balance and top-up from the Account page, where checkout is handled via Creem. You can also enforce a minimum credit balance and set credit budgets for specific runs.