Field noteNov 16, 20266 min read1 source

Datasette Agent 0.6a0 puts an agent prompt on the Datasette homepage and adds hash-mentions for tables and

In one minute

Datasette Agent 0.6a0 puts an agent prompt on the Datasette homepage and adds hash-mentions for tables and queries. For teams exploring data through Datasette, the change makes the agent’s entry point more visible and lets a question refer to a specific data object as it is composed.

What changed: Datasette Agent 0.6a0 adds an agent prompt to the Datasette homepage, including hash-mentions for tables and queries.

What happened

Simon Willison’s announcement describes version 0.6a0 of Datasette Agent, his agent plugin for Datasette that lets users ask questions about their data. The release is framed as an accumulation of small fixes. The most notable change is an agent prompt placed directly on the Datasette homepage, rather than a feature described as requiring users to find a separate destination first.

The homepage example shows a text input and an ask button below the search box, under a “Datasette Agent” heading. While typing “Tell me about #re”, the interface displays autocomplete suggestions for the repos and releases tables. The announcement says the new prompt supports hash-mentions of tables and queries. That makes the intended interaction concrete: a user can begin a question on the homepage and use a short typed reference to select a data object from suggestions.

The release does not describe a change to the underlying agent’s reasoning, data access, or answer generation. Nor does the announcement identify the other small fixes accumulated in 0.6a0. The supported claims are about the user-facing prompt and its hash-mention support; the example shows table suggestions, while the announcement’s description also names queries. It gives no release date beyond the publication date of the announcement, no version history comparison, and no usage or performance measurements.

The practical distinction is between making an existing capability easier to enter and establishing a new capability in what the agent can do. The announcement supports the first claim: the prompt is directly on the homepage, and references can be completed with suggestions. It does not claim that the agent can answer more kinds of questions, that suggestions cover every object, or that the feature changes permissions or data handling. Those questions require testing the release in the relevant Datasette setup.

What the archive adds

New Runtime’s earlier coverage of Datasette 1.0a41 described a move toward shared infrastructure for runtime visibility and plugin-interface consistency. This release is a new event in the same broader product area, but it is not evidence that those earlier infrastructure changes caused the new homepage interaction. New Runtime also covered Datasette Apps 0.3a0, which added permission-aware app references and stored-query controls to workflows using Datasette Agent. The present announcement adds a more visible entry point and typed references, rather than a stated change to those controls.

Trend signal

This release modestly supports agent-ready software exposing capabilities: the homepage presents an explicit agent prompt, and hash-mentions connect questions to named tables and queries. The evidence is about interface design, not a demonstrated permission model or composability across external agents. The release is a small example of making a capability discoverable inside the software where relevant data already lives, rather than proof of a broader product shift.

Why now

The announcement presents the homepage prompt as the most notable item in a release of accumulated fixes, but does not explain why it was added at this point. The timing therefore does not support a claim about a market shift, user demand, or a change in Datasette’s roadmap.

Connections to other events

New Runtime’s earlier coverage of Datasette Apps 0.3a0 noted permission-aware app references and stored-query controls in workflows using Datasette Agent. Together, the events touch different parts of the user experience: controls around references and queries in one release, and a homepage prompt with hash-mention suggestions in this one. The announcement does not say that the features require each other or were developed as a coordinated change.

Why it matters

An agent can be technically present in a product and still be easy to overlook or awkward to invoke. Putting the prompt on the homepage reduces that discovery step for a Datasette user, while hash-mentions offer a lightweight way to point a question at a table or query. That can make exploratory work more specific: instead of starting with an unanchored question, a user can select a relevant object while composing it.

For teams building data tools, the useful product lesson is modest but actionable: the path into an assistant is part of the workflow, and references should be legible at the moment a user needs them. Yet a convenient prompt is not itself evidence of trustworthy answers. Teams still need to evaluate whether the agent interprets references correctly, respects whatever access controls apply in their deployment, and produces results users can verify. The announcement provides no evaluation results on those points.

What happens next

  • Datasette Agent’s homepage prompt will make the agent more discoverable than a separate entry point for users who regularly open Datasette. (3 months, 65%). Watch for: subsequent release notes or product documentation discussing homepage prompt usage or further interaction changes.

Limits and uncertainty

The announcement does not quantify adoption, time saved, answer quality, or the accuracy of hash-mention selection. Its screenshot illustrates suggestions for two tables, but does not establish how suggestions behave across larger datasets or with queries. It also does not explain permissions, data handling, supported model providers, pricing, or the other small fixes in the release. These are not claims that the feature lacks such safeguards or capabilities; they remain questions to check against the release and in a real deployment.

Skill to build now

If your team uses Datasette Agent, test the new homepage prompt with a small set of representative questions and explicit table or query references. Check whether autocomplete surfaces the intended objects, whether the resulting answer stays tied to them, and whether a reviewer can verify the answer against the underlying data. Record failures separately from usability improvements; the announcement describes the interaction, not an evaluation method.

Reader takeaway

Datasette Agent 0.6a0 makes its prompt easier to find and adds hash-mentions for tables and queries. Treat that as an interface improvement, not evidence of stronger answers: try it with real references, then check whether the responses remain accurate and verifiable.

Open question

How does Datasette Agent handle hash-mentions and suggestions when a user has access to only part of a Datasette instance’s data?

Evidence / sources

  1. 1simonwillison.net/2026/Nov/15/datasette-agent
All Field Notes

Next reads

Read next

Open the graph
  1. 01topicAgent InterfacesExplore the agent-interfaces topic hub.
  2. 02topicDeveloper ToolsExplore the developer-tools topic hub.
  3. 03archiveField NotesOpen the latest editorial analysis.
  4. 04sourcesSourcesSee the original sources this archive cites.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract