datasette-agent 0.2a0 adds mid-execution user questions and a save_query tool
datasette-agent 0.2a0 ships two major features: tools can now pause mid-execution to ask the user yes/no, multiple-choice, or free-text questions, and a new built-in `save_query` tool lets the agent save SQL as a Datasette stored query — always requiring human approval before anything is stored.
Score breakdown
The mid-execution `ask_user()` mechanism allows agentic tools to gate side effects on explicit human approval and survive server restarts while awaiting a response, replacing a model where agents had to complete or abort a turn without user input.
- 01datasette-agent 0.2a0 is an alpha release of an LLM-powered agent for Datasette.
- 02Tools can now call `await context.ask_user(...)` to pause mid-execution and ask the user a question.
- 03Three question types are supported: yes/no, multiple-choice (`options=[...]`), and free-text (`free_text=True`).
datasette-agent 0.2a0 is an alpha release of the LLM-powered agent for Datasette. Its headline addition is a mid-execution questioning mechanism: tools can declare a `context` parameter to receive a `ToolContext` object, then call `await context.ask_user(...)` to pause and ask the user a question. Three question types are supported — yes/no, multiple-choice (via `options=[...]`), and free-text (via `free_text=True`). While a question is pending, the agent turn suspends; the question is rendered as a form in the chat UI and persisted to the internal database, meaning suspended conversations survive a server restart. Once the user responds, the tool re-executes from the top with stored answers replayed. The release notes advise calling `ask_user()` before performing any side effects.
The second major addition is a built-in `save_query` tool that allows the agent to save SQL it has written as a Datasette stored query.
The second major addition is a built-in `save_query` tool that allows the agent to save SQL it has written as a Datasette stored query. Human approval is always required: the agent presents the full SQL along with the proposed name, database, and visibility settings, and nothing is stored until the user clicks Yes. The post notes that the `ask_user()` capability was enabled by a new LLM alpha built with the help of Claude Fable 5.
Key facts
- 01datasette-agent 0.2a0 is an alpha release of an LLM-powered agent for Datasette.
- 02Tools can now call `await context.ask_user(...)` to pause mid-execution and ask the user a question.
- 03Three question types are supported: yes/no, multiple-choice (`options=[...]`), and free-text (`free_text=True`).
- 04Suspended agent turns persist to an internal database, so conversations survive a server restart; on resume, the tool re-executes from the top with stored answers replayed.
- 05A new built-in `save_query` tool lets the agent save SQL as a Datasette stored query, but always requires explicit human approval before storing anything.
- 06The `ask_user()` feature was enabled by a new LLM alpha built with the help of Claude Fable 5.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 11, 2026 · 08:34 UTC. How this works →