Gets the work done, on your machine.

Connect the model you already use and give it real tools: files, shell, web, browser. Arabic-first, and every sensitive step waits for you.

Works with the model you already have, from twenty-one providers

AnthropicOpenAIGoogle GeminiDeepSeekMistralxAI (Grok)GitHub CopilotOpenRouterAzure OpenAIAWS BedrockGoogle Vertex AIGroqCerebrasFireworksTogetherQwen (DashScope)Kimi (Moonshot)GLM (Z.ai)OllamaLM StudiovLLM / OpenAI-compatible
How it works

Three steps, then it works on its own

No account, no subscription. Your key, your machine, your folder.

The Models page: five models from different providers, each verified with its latency and a fallback model
A three-point plan turned into three tasks with different statuses
A task page: steps as tool cards, a permission request to run a command, and the git changes panel
What it does

Built to do the work, not just answer

Tasks in parallel

Up to 100 at once. Tasks that touch the same files take turns, tasks that depend on others wait. The reply keeps going even if you leave the page.

A task group inside the chat: two completed and one running

A git worktree per task

On any git project, each task works in an isolated copy. Changes are listed as files and a diff, and one click reverts them.

The changes panel: 26 lines added and 3 removed, applied to the folder, with a revert button

Design with a live preview

Answer a few questions and let the model design with bundled design skills. The preview sits beside the chat at phone, tablet and desktop widths, and one click hands it off to be built.

Cost in view, limits in your hands

Every call is counted in tokens and dollars per model. Set a daily or monthly limit and Rafiq stops before the invoice surprises you.

Spending today and this month, with daily and monthly limits and a thirty-day chart

Twenty-one providers, local models too

From Anthropic and OpenAI to Bedrock, Vertex and Ollama. A fallback model for each, and request limits so you never hit a rate-limit wall.

AnthropicOpenAIGoogle GeminiDeepSeekMistralxAI (Grok)GitHub CopilotOpenRouterAzure OpenAIAWS BedrockGoogle Vertex AIGroqCerebrasFireworksTogetherQwen (DashScope)Kimi (Moonshot)GLM (Z.ai)OllamaLM StudiovLLM / OpenAI-compatible

Keeps working in the background

Close the window and it carries on next to the clock. A notification when a task finishes or needs you, and scheduled tasks daily or every few hours.

  • Starts with Windows if you like
  • Templates for the tasks you repeat
  • Updates itself from inside the app

Any model, and your keys stay yours

Cloud or local: Claude, GPT, Gemini and DeepSeek, or Ollama and LM Studio on your machine. Keys live in the Windows keyring, never in a database.

Skills you write, tasks that run on their own

Write your own instructions for the model inside the app, checked as you type. Let a task run every morning without opening anything.

All features in detail

Privacy

On your machine, with your approval

Keys in the Windows vault

API keys and tokens live in Windows Credential Manager only. They never enter the database, the logs, or the app's interface.

Every sensitive step asks

Writing a file, running a command, browsing, controlling the desktop, calling an MCP tool: each has its own policy. Ask, allow, or deny.

No Rafiq cloud

Your chats, tasks and designs are in a local database. Requests go straight to the provider you chose, and to nobody else.

How Rafiq protects your data

Rafiq's home screen in light mode: finished tasks, today's spending, recent chats and the next scheduled run
Light and dark, and Arabic right-to-left from the ground up. In English and Russian too.
Rafiq's home screen in dark mode

Ready? Half a minute and it's yours.

One installer, no account. Then add a model key and start.

  • SystemWindows 10 or 11, x64
  • Version0.3.0
  • Size87 MB
  • InstallPer user, no administrator rights needed