Introduction

Rinari is an AI engineering companion that works inside your projects. Ask for a change and she reads the repository, plans, edits files, runs commands and tests, and shows you the evidence, so you can review the result instead of piecing it together from terminal output.

One engine, two ways to work

Rinari is three pieces that fit together:

Piece	What it is
Rinari Engine	The runtime. It owns execution, sessions, model routing, tools, approvals, credentials and persistent state.
Rinari CLI	The terminal client, the rinari command. It lives in the same repository as the engine.
Rinari Agent	The desktop app. A visual workspace for projects, tools, agents and results, on top of the same engine.

The desktop app does not reimplement the agent. Both clients talk to the same engine, so a session started in the terminal can continue on the desktop with its history intact. See CLI ↔ desktop.

What you can do

• Open a project or start a chat. Create a project from one or several folders, start a conversation without a project, or resume a saved session.
• Keep separate workspaces. Each profile has its own projects and conversations.
• Choose how she works. Pick PLAN, BUILD or REVIEW; the engine applies the permissions that go with each mode. See Modes.
• Bring context. Attach source files, images, PDFs, Word documents or spreadsheets, and use @ to bring workspace files into the conversation.
• Follow the work. Model messages, tool calls, command output and results appear in an execution timeline, with elapsed time, errors and cancellation state. In multi-step work, a live checklist above the composer shows what’s done and what’s left.
• Stay in control. Answer approvals and questions in context. See Approvals and permissions.
• Review the result. Inspect changes and diffs, tasks, verification evidence, checkpoints, artifacts, context and usage. See Verification.
• Keep execution within reach. The browser panel shows the real Chromium page the engine is using, and the processes panel shows the background commands a session started.
• Talk instead of typing. Dictation runs on your computer; no audio leaves it.
• Let her remember. Rinari’s memory is visible: you approve, review, export or forget what she keeps.
• Work side by side. Boards put several sessions in one window, each with its own project and model. See Boards and Flows.

New in 0.3

Rinari Agent 0.3 brings a new interface, Obsidian Neon, and these features:

Feature	Where to read about it
Profiles as workspaces, also in the terminal with rinari bundles	Profiles
Projects with several folders, each with its own branch and trust	Projects with several folders
A live checklist above the composer	Live checklist
Notes with suggested tasks	Notes
Visible memory you can export and import	Memory
Local dictation with whisper.cpp	Dictation
Full access asks for confirmation	Full access
How much character Rinari shows	Personality
Claude Subscription, experimental and off by default	Connect a provider

Updating from 0.2.x migrates your data; read Install on Windows first.

Bring your own model

You supply the model provider: OpenAI, Anthropic, local servers such as Ollama or LM Studio, any OpenAI-compatible endpoint, and more. Remote providers may charge for usage; local models need a running compatible server and suitable hardware. Tool calling, vision and reasoning options depend on the model you pick. See Connect a provider.

Where Rinari stands

Rinari is in early access and still moving toward a stable release. What a given build includes depends on its version, so check the release notes of the build you install. The desktop app is distributed for Windows x64, and its interface is available in English and Spanish.

Rinari gives you evidence to review a result. A generated answer is not a substitute for tests, and a successful tool call is not a guarantee that a task is correct.

Next steps

1. Install Rinari Agent on Windows.
2. Connect a provider and pick a model.
3. Open your first project.