Your first project

A project is a folder Rinari works in. A conversation without a project is a general chat with its own independent working folder. Both are sessions the engine keeps, so you can close the app and pick them up later.

Open a folder

In the sidebar, choose Open folder. The New project window asks for a name, the folders — one, or several — and whether you trust each one, then Create project. New conversation starts another session in the current project, and New conversation without a project starts a general chat.

Projects and conversations live in the active profile; everything starts in Default.

Decide whether to trust it

In the New project window each folder has its own Trust switch. If you open a project that isn’t trusted, Rinari asks Do you trust this folder? — Trust or Not now.

Trusting a project lets the engine load what the project brings with it: its local instructions and its project-level skills, plugins, MCP servers and hooks. Plugins are code, so only trust folders whose contents you trust. An untrusted project still works; it just runs without those local extensions and shows Untrusted project: local instructions are disabled.

If the project’s Git identity changes after you trusted it — a new remote or a re-created repository — its trust needs to be revalidated, and its local instructions stay disabled until you choose Trust again.

Write your first request

The composer at the bottom holds everything you need for a turn:

• Mode — PLAN, BUILD or REVIEW. See Modes.
• Permissions for this chat — Read-only, Workspace or Full access. See Approvals and permissions; Full access asks for confirmation first.
• Model and reasoning effort, when the model supports it.
• Attachments — source files, images, PDFs, Word documents or spreadsheets. Type @ to search the workspace and add a file.
• Microphone — dictate instead of typing.

A good first request is bounded and says what “done” means. For example:

• Map this repository and propose a plan before editing. (PLAN)
• Implement this change and run the relevant tests. (BUILD)
• Review these changes for regressions. (REVIEW)

These are starting points, not recorded outcomes: results depend on your project, your model, the available tools and your permissions.

Follow the work

Each turn appears as an execution timeline: model messages, tool calls, command output and results, with elapsed time and errors. When Rinari needs a decision, an approval request or a question appears right in the conversation.

In multi-step work, a live checklist above the composer shows what she did and what’s left. If she spots something worth doing outside your request, she leaves a note instead of doing it.

You don’t have to wait for a turn to finish to say something. While she works:

• Enter sends a note she reads after the current step, without stopping.
• Tab queues the message for when she finishes.

The stop button cancels the turn.

Review the result

The workspace panel has a tab for each kind of evidence:

Tab	What it shows
Changes	Files changed and their diffs.
Tasks	The task state Rinari tracks for the work.
Verification	The latest validation evidence and a suggested plan. See Verification.
Checkpoints	Restore points. You can preview a restore before confirming it.
Artifacts	Larger outputs the engine stored during the session.
Context & usage	How much context is in use, and token usage reported or estimated per turn.

The session dock beside the conversation has Files, Browser and Terminal tabs.

Pick it up later

Sessions persist in the engine. Reopen them from the sidebar, or continue the same session from a terminal — see CLI ↔ desktop.