Datasette Agent 0.5a0 Released: Generate Database Apps Through Conversation, Edit in Real Time with Side Panel

Datasette Agent has reached an important version update—0.5a0. This release introduces a core feature called panel, which frees AI conversations from being confined to a chat window and allows them to work side by side with Datasette data pages.

From Chat to Side-by-Side Operation: What Is panel?

The core design idea behind panel is straightforward: once the agent chat window opens, it docks next to the Datasette page—for example, a table page or query page. This way, you can converse with the AI on the left while viewing data or generated apps in real time on the right, forming a "chat, view, and edit as you go" workflow.

This layout solves a pain point of traditional AI conversations: the separation between the conversation and the object being operated on. When you need to refer to specific data to adjust instructions, switching back and forth between windows interrupts your train of thought. panel puts both on the same screen, reducing the cost of context switching.

Integration with datasette-apps: Let Apps "Grow" Through Conversation

The real power of panel is revealed when used together with datasette-apps. According to the official description, you can use a prompt to make a database-driven app appear "from nothing," and then modify it directly in the chat window next to the app.

The official screenshot shows a typical scenario: the user first enters "Build an app to help explore this data," and the agent calls tools such as app_create and app_render to generate the app; then the user says "give it a more hacker aesthetic design," and the agent calls the tools again to modify the style, ultimately presenting a dark terminal-style "LLM Price Explorer" in the right panel. The chat history on the left retains collapsible records of each tool call and version entry points ("View app" and "Edit"), making the iteration process traceable.

This "conversation as development" model further lowers the barrier to app building. You do not need to leave the data page to write code; you only need to describe requirements in natural language, and the agent handles generation and adjustment.

Datasette is an open-source Python data exploration and publishing tool developed by Simon Willison that can instantly turn a SQLite database into a browsable, queryable Web interface with APIs. It does not include AI capabilities itself, but it provides a plugin system (such as datasette-agent and datasette-apps) that makes it an extensible data application platform. In the context of this article, Datasette is the underlying framework that hosts data pages and the agent panel, allowing users to filter, aggregate, and visualize databases without writing backend code. Understanding Datasette's role helps clarify why the agent can directly "see" table structures and query results—because all data operations are exposed to agent tools through Datasette's standardized interfaces.

datasette-apps is a Datasette plugin used to dynamically create, render, and manage small data apps within a Datasette instance. It provides a set of tools (such as app_create for generating an app skeleton based on a prompt or configuration, and app_render for rendering the app as an interactive HTML/JS interface) and maintains app version history. By calling these tools, the agent translates natural language requirements into concrete app definitions, which are then executed by datasette-apps. This division of labor allows the agent to focus on understanding intent and generating configurations, while datasette-apps handles persistence and presentation, enabling iterative modification of app styles and functionality within the conversation.

Hands-On Experience: LLM Price Comparison App

The example app in the screenshot offers a glimpse of how practical this combination is. The right panel displays an app called "LLM Price Explorer," covering 185 models from 12 providers, with workload costs ranging from $0.042 (Jev 1.13) to $300.00 (o1 Pro). The app includes filter controls, workload cost estimation controls, and a table listing matching models with input, output, and cached input prices per million tokens.

This example shows that apps generated by Datasette Agent are not toy demos but interfaces capable of supporting real data exploration tasks. For scenarios that require quickly building internal data tools, this model has practical value.

How to Try It

You can log in to agent.datasette.io with a GitHub account to try it directly. The project is open source on GitHub, with the version tag 0.5a0.

A Brief Observation

The panel feature of Datasette Agent reflects a broader trend: AI assistants are moving from "standalone chat windows" to "embedded workflows." Putting the conversation next to the data, and app generation inside the conversation, may improve the actual user experience more than simply improving model capabilities. For data workers, this means shifting from "asking AI for answers" to "operating on data together with AI."

However, 0.5a0 is still an alpha version, and its features and stability may still be iterating. If you are interested in data exploration and AI-assisted development, this project is worth keeping an eye on.

One-line Summary

Datasette Agent 0.5a0 introduces the panel feature, enabling side-by-side collaboration between chat and data pages.

Paragraph Summary

Datasette Agent 0.5a0 introduces the panel feature, allowing the AI chat window to dock next to Datasette data pages and form a "chat, view, and edit as you go" workflow, solving the pain point of separation between the conversation and the object being operated on. Combined with the datasette-apps plugin, users can generate database apps from nothing through natural language prompts and modify them in real time in the side chat. The official example demonstrates the entire process of building and adjusting an LLM price comparison app through conversation, proving that the generated apps can support real data exploration tasks. This version can be tried at agent.datasette.io with a GitHub account. The project is open source, but it is still an alpha version, with features and stability continuing to iterate.