Plugins

Install a plugin beside SQLsaber by passing the plugin package in --with:

uv tool install --with sqlsaber-viz sqlsaber

The saber CLI loads installed plugins from the sqlsaber.capabilities entry-point group.

An embedded SQLSaber session does not load installed plugins. Import the plugin factories you want and list them in SQLSaberOptions.capabilities. See Capabilities.

Bundled handoff

The handoff plugin ships with SQLsaber and is enabled in the CLI without additional installation or setup. Use /handoff <goal> to generate an editable continuation prompt, or ask the model to draft a handoff using its draft_handoff tool. The tool returns a draft only; it never clears history or starts another thread.

Configure it like any other plugin:

saber plugins show handoff
saber plugins set handoff model openai:gpt-5-mini
saber plugins unset handoff model
saber plugins disable handoff
saber plugins enable handoff

Changes apply to new CLI sessions. Disabling handoff removes both the model tool and draft generation through /handoff; it preserves saved settings.

SQLSABER_HANDOFF_MODEL overrides the saved plugin model. Without a plugin model, handoff uses the active session model. Unsetting the plugin model restores that fallback. Authentication uses the configured provider credentials. When inheriting the session model, it also inherits the session’s explicit API key.

SDK sessions opt in explicitly, including for saber.draft_handoff(). See Bundled handoff.

Visualization

The visualization plugin renders charts in the terminal from query results.

uv tool install --with sqlsaber-viz sqlsaber

If SQLsaber is already installed:

uv tool install --with sqlsaber-viz --force sqlsaber

After a query, ask for a chart:

> Show me the top 10 customers by revenue

| customer_name | total_revenue |
|---------------|---------------|
| Acme Corp     | 50000         |
| Beta Inc      | 45000         |
| ...           | ...           |

> Plot that as a bar chart

You do not need to name the chart type. Describe what you want to see. SQLsaber picks among bar, line, scatter, histogram, and boxplot from the data.

Notebook analysis

The notebook plugin runs multi-step analysis in a sandbox. It uses prior SQL results for calculations, statistics, transformations, and plots. Bounded notebooks and plot previews appear in the terminal. Notebook bytes and images are not sent to the parent model.

uv tool install --with sqlsaber-notebook sqlsaber

Docker is the default local backend. After a query, ask SQLsaber to analyze or plot the results.

Set a dedicated notebook model without changing the main model:

saber plugins set notebook model openai:gpt-5

Select a backend with SQLSABER_NOTEBOOK_BACKEND. Backends never fall back automatically.

Backend	Where it runs	Notes
Docker	Local	Default
Microsandbox	Local microVM	Install the microsandbox extra. Query results stay on the machine.
Modal	Remote	Query results are uploaded to a third party.
Daytona	Remote	Query results are uploaded to the configured Daytona service.

uv tool install --with 'sqlsaber-notebook[microsandbox]' sqlsaber
SQLSABER_NOTEBOOK_BACKEND=microsandbox saber

For an embedded SQLSaber session, list the notebook factory in SQLSaberOptions.capabilities. Persist notebooks and plots with SQLSaberOptions.artifact_store. See Results and streaming.

Python sandbox

The sandbox plugin runs Python analysis and keeps variables, installed packages, and generated files available for follow-up requests. An analysis agent writes and runs the code. SQLsaber receives the findings and artifact references.

To use E2B, set E2B_API_KEY and install the plugin:

uv tool install --with 'sqlsaber-sandbox[e2b]' sqlsaber
SQLSABER_SANDBOX_PROVIDER=e2b saber

After a query, ask SQLsaber to analyze the results, save a plot, or train a model. For a follow-up, ask it to continue the same analysis.

To use another provider, install its extra and set SQLSABER_SANDBOX_PROVIDER. Docker requires no extra. It needs the Docker CLI and a running daemon. For microsandbox on Linux, provide read and write access to /dev/kvm.

See Sandbox providers for extras and credentials. For embedded use, configure the capability with SandboxConfig as shown in Persistent sandbox analysis.

Install several plugins

uv tool install --with sqlsaber-viz,sqlsaber-notebook,sqlsaber-sandbox sqlsaber

Build a plugin

Export a capability factory through the sqlsaber.capabilities entry-point group. See Capabilities for the factory contract, PluginContext, and display-tool registration.

The former sqlsaber.tools entry-point group and global ToolRegistry no longer load. Keep your execution function or Tool class, wrap it in a capability, and export capability(context) through sqlsaber.capabilities.