> ## Documentation Index
> Fetch the complete documentation index at: https://docs.lilfella.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Fella

> Ask questions about your own files and inspect the work behind tool-backed answers.

# Ask your files, not a black box

Fella is a local-first desktop app for personal analytics. Open a folder of statements, exports, logs, notes, or other supported files, ask a question in plain language, and inspect the tools and results used to answer it.

<CardGroup cols={3}>
  <Card title="Get an answer" icon="play" href="/getting-started">
    Install Fella, open a folder, and inspect your first tool-backed result.
  </Card>

  <Card title="Try a recipe" icon="list-checks" href="/examples">
    Copy prompts for comparisons, joins, document lookup, and charts.
  </Card>

  <Card title="Fix a problem" icon="wrench" href="/troubleshooting">
    Diagnose missing files, model errors, Python failures, and weak answers.
  </Card>

  <Card title="Check capabilities" icon="table-properties" href="/getting-started#what-the-shipped-app-can-use">
    See the exact shipped formats, providers, tools, and outputs.
  </Card>

  <Card title="Know the limits" icon="triangle-alert" href="/limitations">
    Understand verification, security, format, and scale boundaries.
  </Card>

  <Card title="Understand the boundary" icon="shield-check" href="/extensions-and-mcp">
    See why the personal release keeps tools fixed and MCP experimental.
  </Card>
</CardGroup>

## The useful distinction

| Layer            | Responsibility                                               | What it does not guarantee                             |
| ---------------- | ------------------------------------------------------------ | ------------------------------------------------------ |
| Model            | Interprets the question and chooses tools                    | That its interpretation is correct                     |
| Fella harness    | Supplies workspace context and runs the tool loop            | That every response needs or contains evidence         |
| Analytics engine | Runs read-only SQL, document reads, charts, or Python        | That a valid computation answers the intended question |
| Verification     | Replays queries and checks grounding and known risk patterns | Semantic correctness for every answer                  |

Fella can answer a general question without calling a tool. Those answers have no workspace evidence. When tools run, the working fold records the selected files, arguments, SQL or Python, result samples, timings, errors, and post-answer checks.

<Note>
  The agent has no file write, move, or delete tool. The user can edit the
  `fella.md` context file, while generated Python runs inside an embedded WASM
  capability boundary. Review the [limits](/limitations#security-boundaries)
  before using sensitive data.
</Note>

## What you can do

| Outcome                                  | Typical path                                                                                    |
| ---------------------------------------- | ----------------------------------------------------------------------------------------------- |
| Compare periods or categories            | Ask a plain-language question; Fella queries one or more tables                                 |
| Find a fact in notes or a text-layer PDF | Fella searches and reads document text directly                                                 |
| Reconcile several exports                | Name the files and join key when you know them                                                  |
| Produce a bar or line chart              | Ask for a chart; category charts are bounded, while time-series charts can cover longer periods |
| Inspect data directly                    | Use `/files`, `/schema <table>`, or `/sql <query>`                                              |

```text filename="first question" theme={null}
/open /path/to/your/files
Compare monthly dining spend in the first and second half of 2025. Show the totals and the change.
```

## Local-first, with explicit egress

The provider you connect receives the prompt and tool results needed for the question. `/update` contacts GitHub only when invoked. `/mcp` is inert in the personal release, and generated Python stays inside its embedded local guest with no network capability.

## Next steps

<CardGroup cols={2}>
  <Card title="Start in five steps" icon="route" href="/getting-started">
    Reach a useful, inspectable result from a fresh install.
  </Card>

  <Card title="Understand evidence" icon="search-check" href="/developer-platform/using-fella/evidence">
    Learn what the working fold proves and what you still need to review.
  </Card>
</CardGroup>