Skip to content
Guides
Working with AI

Working with AI

  • Add AI to your app: Use a language model in your app: summarize text into a typed struct, answer questions with tools that look up the user’s own data, and stream the answer as it’s written. Package ai does the work around the model: schemas, validation, the tool loop, logging and tests that never call a model. The complete example is examples/ai, a small support desk API.
  • Build an AI assistant: Give your users a chat with an agent that answers from your app’s data: conversations stored in the database, answers streamed to the page as they’re written, slow questions answered in the background, and a usage budget per user.
  • Search by meaning: Find records by what they mean, not only by the words they share: “how much is the team plan” finds the article on plans and billing, which never says “how much”. Each record’s text is split into chunks, turned into vectors (embeddings) by an embedding model, and stored next to the record; a search embeds the question and returns the nearest records, with the passage that matched. With a full-text index too, the search is hybrid: records that its words find rank high as well, so names, codes and rare words aren’t missed. An agent can use the search as a tool, to answer from your data (retrieval-augmented generation).