Datasette: WordPress for Data

"Datasette is the fastest way to publish data online as an interactive,
searchable database." -- Simon Willison

Datasette takes any SQLite database and gives it a web UI, JSON API,
full-text search, faceting, filtering, CSV export, and 154+ plugins.
One command. Zero backend code.

cursor-mirror already produces SQLite databases (transcript indexes,
consolidated exports). Datasette is how those databases get a face.

Quick Start

# Install (in cursor-mirror venv or globally)
pip install datasette

# Serve Cursor's own database (read-only, safe)
datasette ~/Library/Application\ Support/Cursor/User/globalStorage/state.vscdb

# Serve a cursor-mirror consolidated export
python -c "
import sys; sys.path.insert(0,'.')
from lib.datasette_export import export_datasette
from pathlib import Path
export_datasette(Path('cursor-mirror.db'))
"
datasette cursor-mirror.db -o

The -o flag opens your browser. Every page has a JSON equivalent (add .json to the URL).

What You Get

Table	Rows (typical)	What
composers	200+	Every conversation with metadata, bubble counts
transcript_sections	30,000+	Parsed sections: USER, ASSISTANT, TOOL_CALL, THINKING
shell_commands	2,000+	Every terminal command with line numbers
tool_calls	2,000+	Tool usage with params, error status
feature_flags	30+	Model vs live flags with drift detection

Full-text search is enabled on transcript previews. Faceting works on
section type, tool name, composer, workspace.

Key Datasette Features

URLs are the API. Every query, filter, and facet combination has a
permanent URL. Share it with a colleague. Add .json for machine access.
Add .csv for spreadsheet export. This is Datasette's core insight.

Faceting. Click a column name to see value distribution. Filter by
clicking values. Combine multiple facets. No SQL needed.

SQL. Full SQL editor built in. Write any query, get results as HTML,
JSON, or CSV. Canned queries can be pre-configured.

Plugins. 154+ plugins for charts (datasette-vega), maps
(datasette-cluster-map), dashboards (datasette-dashboards), auth,
search, and more.

Publishing. datasette publish vercel cursor-mirror.db deploys
your data to a public URL in one command. Also supports Fly, Cloud Run.

Integration with cursor-mirror

Direct serve (zero code)

Datasette can open any of cursor-mirror's SQLite files directly:

# Cursor's global database (ItemTable, cursorDiskKV)
datasette state.vscdb

# A transcript index (sections, shell_commands, tool_params)
datasette ~/.cursor/projects/*/agent-transcripts/*.db

# Multiple databases at once
datasette state.vscdb transcript-index.db

Consolidated export

The export_datasette() function in cursor-mirror/lib/datasette_export.py
produces a single optimized database combining:

• All composer metadata
• Parsed transcript sections (with FTS)
• Shell commands
• Tool calls from bubbles
• Feature flag drift

Planned: datasette-cursor-mirror plugin

A Datasette plugin that auto-discovers cursor-mirror data, renders
transcript sections as formatted HTML, and adds a dashboard. See
designs/DATASETTE-CURSOR-MIRROR-INTEGRATION.md for the full plan.

Useful Datasette Commands

# Serve with CORS (for JavaScript access from other domains)
datasette cursor-mirror.db --cors

# Serve read-only (immutable mode, enables caching)
datasette -i cursor-mirror.db

# Install a plugin
datasette install datasette-vega

# Publish to Vercel (free tier)
datasette publish vercel cursor-mirror.db --project my-cursor-data

# Run a SQL query from the command line (no server)
datasette cursor-mirror.db --get '/cursor-mirror.json?sql=select+tool_name,count(*)+from+tool_calls+group+by+1+order+by+2+desc&_shape=array'

Useful SQL Queries (Canned Queries for Datasette)

Most used tools

select tool_name, count(*) as calls, sum(is_error) as errors
from tool_calls group by tool_name order by calls desc

Shell commands by workspace

select workspace, count(*) as commands
from shell_commands group by workspace order by commands desc

Feature flag drift

select flag, model_enabled, live_enabled
from feature_flags where drift = 1 order by flag

Busiest composers

select c.name, c.bubble_count, c.workspace_folder,
       (select count(*) from transcript_sections ts where ts.composer_id = c.id) as sections
from composers c order by c.bubble_count desc limit 20

Thinking blocks (search for keywords)

select composer_id, start_line, end_line, text_preview
from transcript_sections where type = 'THINKING'
and text_preview like '%security%' order by start_line

About Datasette

Created by Simon Willison (co-creator of Django). Open source, Apache 2.0.
10.8k GitHub stars, 154+ plugins, Python 3.14 compatible.

• Website: https://datasette.io
• Docs: https://docs.datasette.io
• Plugins: https://datasette.io/plugins
• Source: https://github.com/simonw/datasette
• Cloud: https://www.datasette.cloud

Part of MOOLLM

This is a MOOLLM skill. See the MOOLLM README and
skills index for context.