Datasette Apps 0.4a0 Released: Inter-App Linking and Agent Integration

datasette-apps 0.4a0 enables sandboxed apps to interlink via URL fragments and exposes link schemes to the Agent.
datasette-apps are sandboxed HTML/JavaScript apps running inside Datasette that can execute SQL queries directly against Datasette-managed data—similar to Claude Artifacts but with data query capabilities. Version 0.4a0 focuses on "linking" with two changes: support for #fragments in URLs to link to apps and allow apps to read and use these fragments; and reporting app-supported URL schemes back to the Datasette Agent so it can generate code that links apps together. Together these transform apps from isolated pages into composable nodes, with the typical scenario being search apps and detail apps separated and connected via links, achieving separation of concerns and reusability. The article also notes that this version is still alpha and interfaces may change, so production use should first be validated on a small scale.
What Are Datasette Apps
Datasette Apps are sandboxed HTML and JavaScript applications that run inside Datasette. The official documentation compares them to Claude Artifacts, but with a key difference: these apps can execute SQL queries directly against data managed in Datasette. In other words, it puts a "lightweight frontend" and "data query capabilities" into the same sandbox environment, letting users quickly assemble interactive micro-tools around their data without standing up a separate service.
This design continues Datasette's long-standing philosophy—opening up data while lowering the barrier to consuming it. Developers can build interfaces with familiar HTML/JS, while SQL queries are handled on the Datasette side, with the two ends connected through agreed-upon interfaces.
Datasette is a Python tool developed by Simon Willison for publishing SQLite databases as browsable, queryable web interfaces. Its core capabilities include auto-generated table views, support for arbitrary read-only SQL queries, and a JSON API. Its plugin system allows extensions for rendering, authentication, and export functionality.
Sandboxing here means the app runs in a restricted HTML/JS execution environment that cannot arbitrarily access the host page or arbitrary network resources—it can only issue data queries through bridge interfaces explicitly provided by Datasette. This design constrains the attack surface while opening up interaction capabilities, and it's also the premise for understanding why datasette-apps needs "agreed-upon interfaces" and why link schemes must be explicitly reported back to the Agent.
The Core Change in 0.4a0: Links Become First-Class Citizens
#fragments Support
The new version allows using #fragments in URLs to link to an app, and the app itself can read and use these fragments. This means an app is no longer just a page that gets opened—it can receive parameterized context passed in from outside.
URL fragments (#fragment) are the part of a URL after #. Per the HTTP specification, they are not sent to the server and are typically read and interpreted by browser-side scripts. In single-page applications and sandboxed app scenarios, fragments serve as a lightweight parameter-passing channel: callers can encode context such as "which record to display" or "what filter conditions to use" into a link without changing server-side routing, and the app's own JavaScript parses it to decide what to render.
Therefore, allowing apps to read #fragments essentially defines a "deep link" entry point for each app that can be constructed externally, enabling references between apps without requiring server-side changes.
URL Schemes Reported Back to Datasette Agent
More noteworthy is the second point: the URL schemes supported by apps are reported back to the Datasette Agent. The Agent therefore knows which apps can be linked, and in what form, so that when helping users write apps, it can generate code that links them together.
Combined, these two capabilities directly change how apps are organized. Previously, each app was more like an isolated island, and users had to manually switch between apps; now apps can establish explicit reference relationships.
Datasette Agent is a conversational assistant for Datasette that can generate app code or queries based on natural language requests. The "URL scheme" here does not refer to protocol prefixes like https:, but rather the linking conventions an app exposes externally—which fragment parameters it accepts, their meanings and value ranges, and which app path they point to.
Reporting these conventions back to the Agent is equivalent to providing it with a searchable "app interface catalog." When generating new apps, the Agent can use this to write correct inter-linking code rather than guessing at invalid addresses. This also explains the pragmatic integration approach emphasized in the article: the Agent understands interface contracts, rather than directly touching data or the runtime environment.
Typical Scenarios for Composable Apps
The official example is straightforward: one app displays detailed information about a record, and another completely independent app handles searching for these records and links to the corresponding display page in the first app.
This separation brings several practical benefits:
- Separation of concerns: Search and detail views evolve independently without being crammed into the same app;
- Reusability: A detail app can be referenced by multiple search or list-type apps;
- Agent-friendly: Because URL schemes are visible to the Agent, auto-generated apps naturally support inter-linking rather than producing a bunch of disconnected black-box pages.
From a product perspective, this approaches an "app-as-route" philosophy—each app has stable address semantics, and composition relationships are expressed through links rather than code coupling.
Significance for the Datasette Ecosystem
Datasette has long positioned itself around "publishing and exploring data," and datasette-apps fills in the interaction layer. The linking capabilities in 0.4a0 give this layer network effects: as the number of apps grows, inter-linking increases overall value rather than merely producing a long list of parallel entry points.
At the same time, exposing URL schemes to the Datasette Agent reflects a pragmatic approach to Agent integration—not letting the Agent directly manipulate data, but letting it understand app interface conventions and generate code that conforms to them. This is consistent with the overall sandboxing design: capabilities are open, but boundaries are clear.
Note that this version is still in alpha (0.4a0), and interfaces and behavior may continue to change before the official release. For production environments, it's advisable to first validate link semantics and the stability of Agent-generated results on a small scale.
Getting Started Suggestions
If you're already using Datasette and have tried datasette-apps, you can start from two directions: first, refactor existing detail-type pages into forms that can be parameterized via #fragments; second, involve the Datasette Agent in generating inter-linked apps and observe whether its understanding of URL schemes is accurate. For developers who haven't yet explored this, this version also provides a good window into understanding the combined pattern of "sandboxed apps + data queries + Agent collaboration."
Related tags: datasette-agent, datasette-apps, datasette, urls.
Related articles

RePlan-VLA: Progress-Aware Replanning Breakthrough for Long-Horizon Robotic Manipulation
The RePlan-VLA framework converts VLA models' short-horizon capabilities into reliable long-horizon execution through progress-aware adaptive replanning, improving success rates by 21.9% and 10.4% on RoboCasa365 and CALVIN, and validating real-world deployment on a Franka Research 3 robot.

Geometric Bridge Flow Matching: A New Breakthrough in Blind Rigid Motion Correction
Geometric bridge flow matching enables blind rigid motion correction in head CBCT, achieving 37.1 dB PSNR and 0.27 mm RPE on simulated data in ~9.1 minutes, outperforming existing methods.

Google Releases Gemini 4 Hydrogen: 2.5x Faster Without Sacrificing Intelligence
Google releases the lightweight Gemini 4 Hydrogen: 2.5x faster task completion, AA Intelligence Index of 54 beating its flagship sibling, 7.3s single-step on Agent Arena faster than Claude and GPT rivals, four thinking levels with pay-as-you-need pricing, now available across AI Studio and more.