A clearer desktop workspace
- The left sidebar is organized around Chat, Files, and Reports.
- A mounted folder is selected from the title bar and remains the explicit source boundary.
- Notebook can sit beside Chat to show the active Investigation. Its panes adapt to narrower windows, and the app retains native operating-system window controls.
- The file view includes a folder hierarchy, catalog status, and supported source previews.
- The shell/terminal feature has been removed from the product.
Workspace knowledge
The Workspace context view brings the user-authoredfella.md guide together with reusable knowledge derived from analysis and Investigation definitions. Learned facts keep their provenance, evidence, freshness, and scope. The user can keep a proposal within its Investigation, edit it, share it across the workspace, or dismiss it; Investigation-local definitions do not silently become workspace-wide rules.
Persistent Investigations
Investigations are now durable, named records with a linked conversation. They can hold an editable brief, an ordered Question plan, per-Question source choices, definitions and assumptions, answer status, saved runs, and Findings. Each run records its workspace revision, actual sources, context snapshot, execution trace, results, and verification. Results distinguish answered work from incomplete, unsupported, blocked, outdated, or still-unstarted Questions. A completed run does not automatically mean the whole Investigation has complete coverage.Result-bound generative presentations
The default structured-answer path now uses JSON-render. The model can compose prose, cards, metrics, tables, charts, and supported interactive controls in a response-specific layout. A chart is used when it explains the result better than a table; the system does not force every result into one presentation form. Generated presentations reference saved analytical results. The renderer validates the component tree, fields, and bindings and does not execute model-authored code. A presentation failure leaves the canonical answer and analysis available as text. Saved presentations can be restored without rerunning the analysis.Reusable results and reports
- Analytical outputs have workspace-scoped identities and revisions tied to the run that produced them.
- Users can revise a saved result, compare its meaning, inputs, method, results, and known report impact, then explicitly adopt a revision.
- Reports can combine authored narrative with exact pinned output revisions.
- Reports can be exported as Markdown or as a portable Fella snapshot. Imported snapshots are read-only and cannot query the original files.
- Saved outputs and their supporting analysis records survive deletion of their conversation; the transcript and linked Investigation are deleted.
Model choice and app navigation
Provider credentials remain user supplied. A conversation’s selected model is visible and changeable from the composer; the current app does not automatically route between specialist models. The existing provider options and defaults are listed in Models and providers. The guide pages now follow the current product model: Chat is the conversational surface, Investigations own persistent analysis, and Reports preserve selected work for review and handoff.Product model
Understand the current objects and how they relate.
Investigations
Use the Brief, Sources, Questions, Analysis, and Results views.
Reports
Save, revise, compare, and export analytical outputs.
Getting started
Set up a provider and open a folder in the desktop app.