datasette-agent adds mid-execution user questions (0.2a0)
TL;DR
datasette-agent 0.2a0 lets tools ask yes/no, multiple-choice or free-text questions with context.ask_user. Unanswered questions suspend execution and persist in the database across restarts.
What changed
The datasette-agent 0.2a0 release adds support for tools to ask questions during execution. A ToolContext object enables yes or no, multiple choice with options, or free text questions via context.ask_user. Conversations stay suspended until answered and survive server restarts through database persistence.
Why it matters
Developers building agent based data tools benefit from this interactive capability which supports three question formats in ongoing sessions. Basic users experience fewer disruptions in chat based queries compared to static agent runs in frameworks like LangChain.
What to watch for
Vibe builders should test against alternatives such as CrewAI for similar question flows. Check the persisted forms in the internal database after a server restart to confirm the feature works.
Who this matters for
- Vibe Builders: Test the new interactive question flows against CrewAI to improve agentic data workflows.
- Basic Users: Use the chat UI forms to guide agents through complex data queries without losing session state.
Harsh’s take
Simon Willison is solving the biggest friction point in agentic workflows: the black box problem. Most agents fail because they make assumptions instead of asking for clarification. By persisting questions to a database, datasette-agent ensures that a server restart does not kill a complex task.
This is a pragmatic shift away from the fire-and-forget model of early LLM apps. Operators should notice the move toward human-in-the-loop as a core architectural feature rather than a UI afterthought. Using a ToolContext object to handle stateful interruptions is a clean pattern for anyone building data-heavy tools.
It moves the needle from simple chatbots to reliable software that can handle ambiguity without crashing or hallucinating a wrong path.
by Harsh Desai
More AI news
- FeatureLius model applies continual instruction tuning for Kupang Malay translation
Lius introduces an LLM fine-tuned via continual instruction tuning to improve translation for low-resource Kupang Malay.
- FeatureBenchmark frames hour-long video grounding as search problem
New benchmark and decomposition examine natural-language temporal grounding over hour-long videos, extending prior work limited to short clips.
- FeatureOn the Limits of LLM-as-Judge for Scientific Novelty Assessment
LLMs now generate and judge scientific ideas, making novelty evaluation a key challenge. Researchers examine research questions as a focused case separate from full method and feasibility assessment.