AI Assistant

Ask questions, get writing help, and take action on your briefs — all from the conversation panel alongside your work.

Overview

The AI assistant is available in every brief, blueprint, and method conversation in Hamster Studio. You interact with it by typing in the chat panel. The assistant understands the document you're currently working on, the team knowledge your workspace has built up, and the context you've explicitly attached. It can answer questions, write and update document sections, and trigger actions like generating a task plan — responding immediately in the same thread where you're working.

When your request involves research, file analysis, or URL extraction, the assistant handles it automatically — you don't need to choose anything. You see progress steps as they run.

How It Works

  1. Open the chat panel — On any brief, blueprint, or method page, the AI chat panel is available in the sidebar. Type your message and press Enter or click Send.

  2. The assistant reads your context — Before responding, the assistant considers what document you have open, any files or URLs you've attached as context, and relevant knowledge from your workspace (blueprints, methods, previous conversations).

  3. Requests are routed automatically — Simple questions get answered directly. Requests involving research, file analysis, or URL extraction are handed to specialist agents. You see a "Working…" indicator with each step as it completes.

  4. The response is delivered in-thread — The assistant's answer appears in the conversation. If it updated your document, the changes appear in the editor immediately. You can continue the conversation naturally from there.

Key Capabilities

  • Document editing: Ask the assistant to add a section, rewrite a paragraph, or restructure your brief. Changes are applied directly to the open document without copy-pasting.

  • In-context Q&A: Ask questions about your brief's content, your team's guidelines (blueprints), or your process library (methods). The assistant searches your workspace knowledge base before answering.

  • Multi-step flows: Structured flows handle more complex requests — researching a topic from multiple sources, extracting content from a URL, processing meeting notes — and run as a visible sequence of steps in the chat.

  • Try again: If a response misses the mark, you can ask the assistant to try again from the same message. It re-runs the last flow with any additional guidance you provide.

  • Replay: The assistant remembers your last query in the current thread. You can replay it after making changes to your document to see how the answer updates.

  • One-tap suggestions: When you open a new conversation, Hamster surfaces quick-start prompts organized into categories (ideas, strategy, features, code). These help you start writing without staring at a blank input. This feature may need to be enabled for your workspace.

How the Assistant Thinks

When your message arrives, the assistant decides how to handle it:

  • Direct response — If your message is a question or a conversational exchange, the assistant responds immediately without calling any tools.
  • Document action — If you ask the assistant to write, update, or restructure your document, it calls the appropriate editing tools and applies changes in place.
  • Flow execution — If your request requires research, URL extraction, file analysis, or a combination, the assistant triggers a multi-step flow. You see each step labelled in the chat as it runs (for example: "Searching organizational context", "Researching codebase", "Updating brief").

During a flow, the assistant collects results from parallel steps, synthesizes them, and then produces the final output — whether that's a written addition to your document or a standalone answer in the thread.

Tips

  • Be specific about what you want the assistant to do to your document. "Add a requirements section covering authentication, data storage, and API rate limits" produces better results than "update the brief."
  • Attach relevant files or URLs to your brief's context before asking the assistant to incorporate them. See Document Analysis for how context documents work.
  • If the assistant produces a result that's close but not quite right, continue the conversation rather than starting over. It retains the thread history and can refine its previous response.
  • For structured research across your codebase, organizational docs, and external sources, explicitly ask the assistant to "research" the topic. This triggers the full multi-source research flow.
  • The assistant works within your workspace's knowledge — the more complete your blueprints and methods are, the more context it has to work with.

Related