Save and rerun results¶

datasight offers three ways to return to past work: bookmarks (reusable prompts), reports (saved SQL that re-runs without the AI), and conversation history (full chat replays).

Bookmark a query¶

To save a query for later reuse, hover over any result and click ★ Bookmark. You can also bookmark from the query history panel (the ★ button on each query card).

Bookmarks appear in the Bookmarks sidebar section. Click a bookmark to populate the chat input with its SQL. Bookmarks are stored in .datasight/bookmarks.json and persist across server restarts.

To clear all bookmarks, click Clear in the Bookmarks section header.

Save a report¶

Reports save a query and re-run it against fresh data without involving the AI. Unlike bookmarks (which only populate the chat input), reports execute immediately and display the result inline.

Save¶

Hover over any result (table or chart) and click Save Report. The report captures the SQL, tool type, chart specification (if applicable), and the title.

Run¶

Click a report in the Reports sidebar section. datasight re-executes the saved SQL and renders the result directly in the chat — no LLM call, no token cost.

Manage¶

Delete — click the x button on any report in the sidebar
Clear all — click Clear in the Reports section header

Reports are stored in .datasight/reports.json and persist across server restarts.

Run reports from the CLI¶

datasight report list
datasight report run 1
datasight report run 1 --format csv
datasight report run 2 --chart-format html -o trend.html
datasight report delete 1

Conversation history¶

Chat conversations are saved automatically and persist across page reloads and server restarts. They're stored as JSON files in .datasight/conversations/.

The History sidebar section lists past conversations. Click one to switch to it — the full chat (messages, tool results, charts) is replayed, and you can continue where you left off.

New Chat in the header clears the current chat screen, SQL history, and dashboard, then starts a fresh working session. Saved conversations remain available in History. The N keyboard shortcut does the same thing.

To clear all history, click Clear in the History section header.

Export a conversation as HTML¶

Click the export button (download icon) in the header to enter export mode. Each question gets a trash button — click it to exclude that Q&A turn (question, SQL, results, and answer). Then click Export HTML to download a self-contained HTML page.

You can also export from the command line:

# List available sessions
datasight export --list-sessions

# Export a session
datasight export <session-id> -o my-analysis.html

# Exclude specific turns by index (0-based)
datasight export <session-id> --exclude 2,3

Export a session as a Python script¶

When you want a runnable, hand-editable record of an analysis — to audit the SQL, share with colleagues who don't run datasight, or wire into a pipeline — export the conversation as a Python script. From export mode, click Export Python script instead of Export HTML.

The downloaded datasight-session.py contains:

A short docstring with the session title and what each section means.
The project's database path baked in as DEFAULT_DB_PATH, with a --db flag to override at runtime.
One labelled section per turn — the user question as a # ─── Turn N ─── header, the SQL as an editable SQL_N = """...""" constant, the chart spec as CHART_N_SPEC = json.loads(...) fed into a Plotly go.Figure, and the assistant's narrative preserved as # Assistant: ... comments.

Run the script standalone:

python datasight-session.py
python datasight-session.py --db /path/to/other.duckdb --output-dir charts/
python datasight-session.py --help

Edit any SQL_N constant to tweak filters, group-bys, or aggregations, then re-run — no AI in the loop, no datasight installation required (only duckdb/sqlite3, pandas, and plotly). Charts are written to the current directory by default; use --output-dir to redirect.

You can also export from the CLI:

datasight export <session-id> --format py -o my-analysis.py
datasight export <session-id> --format bundle -o my-analysis.zip
datasight export <session-id> --format bundle --include html,sql,python,csv,charts,metadata

Bundle exports package the session into a portable zip archive with a versioned manifest.json. Depending on --include, the archive can contain the HTML report, runnable Python script, SQL scripts, CSV result extracts, Plotly chart specs, and structured metadata for provenance and reruns.

DuckDB and SQLite sessions get a fully-runnable connection block. Sessions backed by PostgreSQL or Flight SQL produce a script with a clearly-marked run_sql scaffold for you to wire up your own driver — the script still parses and shows you exactly where the change goes.