Before you start
- Open a folder containing a supported table or document format.
- Connect a model provider with your own API key.
The four-part workflow
1
Open a workspace
Choose a folder, drag it into the app, or use
/open. Fella catalogs supported files and prepares tables and documents for read-only tools.2
Ask a question
Ask about tables, documents, trends, comparisons, joins, or correlations. SQL is the default computation path; Python is the escape hatch.
3
Inspect the evidence
Open Evidence to see tool steps, queries or arguments, results, timings, and verification findings. Charts render below the answer.
4
Refine the workspace
Ask a follow-up, add
fella.md context, correct vocabulary, switch models, or open another conversation.Choose a guide
Open a folder
Workspace boundaries, scanning,
.fellaignore, and the catalog.Ask questions
How to phrase questions and how Fella chooses SQL, documents, Python, and charts.
Read the evidence
Follow every computation and understand the verification checklist.
Understand files
Supported formats, table names, documents, and rescanning.
Add context and memory
Teach Fella your vocabulary without turning context into fake evidence.
Choose a model
Hosted providers, custom endpoints, and credential storage.
Manage sessions
Tabs, the history sidebar, focus mode, retrying, and updates.
Commands and shortcuts
The command palette, slash commands, and keyboard reference.
Product boundary
The base agent has no tool that writes, moves, or deletes workspace files. The model chooses read-only tools; the data engine runs their computations and records evidence. Verification checks the completed answer, but it does not prove that a valid query interpreted an ambiguous question correctly.Fella answers questions about a workspace. It is not a general task agent, file manager, coding assistant, or cloud service.
The harness, engine, and model
The ownership boundaries behind the user workflow.
Privacy and security
Provider egress, credentials, and the Python boundary.