Simon Willison's AI-Augmented Datasette Ecosystem: Agent, Apps, and Plugins

Simon Willison is building AI-native Datasette plugins (datasette-agent [1], datasette-apps [3], datasette-agent-charts [2]) while advancing Datasette core: version 1.0a35 adds Create and Alter table interfaces backed by a JSON API and formalizes a stable template context API for custom templates until Datasette 2.0 [12]. sqlite-utils 4.0rc1 adds a migrations system and nested transaction support [11]. Willison is exploring browser-native persistent SQLite editing in Datasette Lite via the Origin Private File System [13], and used a mixed Claude Code + Codex Desktop (GPT-5.5) workflow to build a browsable Mozilla browser compatibility database in Datasette Lite [9].

The project is a running case study of AI-augmented solo development covering security trade-offs, vibe coding productivity vs. understanding, and now mixed-AI tool division of labor. The potential addition of browser-native persistent SQLite editing to Datasette Lite would extend Datasette's reach beyond server-dependent deployments.

Simon Willison has been building Datasette — an open-source tool for exploring and publishing SQLite databases — since 2017. In 2026 he has layered AI capabilities into the ecosystem through a set of LLM-powered plugins, while using AI tools extensively as development, security, and infrastructure instruments.

The AI-native plugin set includes four components. datasette-agent is a chat interface letting an LLM interact with Datasette through defined tools; version 0.3a0 added an execute_write_sql tool that prompts for user approval before any database modification, with an --unsafe flag for power users who want to bypass approvals entirely [1]. datasette-agent-charts extends the agent with chart generation [2]. datasette-apps provides sandboxed HTML+JavaScript applications inside iframes; apps interact with the database only through pre-approved parameterized stored queries, and CSP configuration is restricted to users holding a dedicated apps-set-csp permission, with an optional allowed_csp_origins allow-list for non-privileged users [3][4]. A security evaluation using Claude Fable 5 found a path where a low-privileged user could exfiltrate data from an administrator who visited a malicious app; Willison fixed the vulnerability before public release [3]. datasette-tailscale arrived at 0.1a0 using Python bindings for the experimental tailscale-rs Rust library to expose a local Datasette server over a Tailnet [5].

Willison uses multiple AI tools across the project. Claude Code (Opus 4.8) has been applied to implementation research — including a three-approach investigation of SQLite column provenance mapping [6] — infrastructure work via Cloudflare WAF configuration [7], and vibe coding: a full port of the Moebius 0.2B image inpainting model to browser WebGPU where Willison read none of the generated code [8]. He acknowledged the trade-off directly: delegating all code generation without reviewing output means not learning the underlying technologies. A browser-compat-db project demonstrated a mixed-AI workflow where Claude Code generated a Python conversion script and Codex Desktop (GPT-5.5) built the GitHub Actions workflow, producing a SQLite database of Mozilla browser compatibility data browsable in Datasette Lite [9]. He has also amplified the argument that MCP's core value is isolating auth flows outside the agent context window rather than integrating tools or skills [10].

Datasette core has advanced alongside the plugin work. sqlite-utils 4.0rc1 ported the sqlite-migrate package's migrations system into the library and added a db.atomic() context manager for nested SQLite transactions using savepoints [11]. Datasette 1.0a35 introduced Create and Alter table interfaces backed by a JSON API — covering columns, primary keys, types, NOT NULL constraints, defaults, and single-column foreign keys — and committed to a stable template context API for custom templates until Datasette 2.0 [12]. Willison is also investigating whether Datasette Lite, which runs the full Python Datasette application in the browser via Pyodide and WebAssembly, might support persistent SQLite editing using the Origin Private File System [13].