Features
Everything Rafeeq can do
All of it runs on-device. Nothing here needs — or uses — an internet connection.
Local chat
Prompt in, streamed response out, powered by a small quantized model running on CPU. No account, no rate limits, no usage sent anywhere. Set the default temperature, response length and CPU threads in Settings.
Model manager
Point Rafeeq at a sideloaded .gguf file, swap models, and see size and quantization info at a glance. Save named profiles for quick switching. Chat and embedding models are labelled automatically.
Code-help mode
Paste or select code, ask questions, get explanations or suggestions. Chat and explain only — Rafeeq never executes code.
Document Q&A & summarization
Load a local .txt, .md, .docx, or .pdf file and ask questions about its contents or get a summary. Long documents are truncated to fit the model's context.
Local RAG
Index a folder of documents with a sideloaded embedding model, then ask questions answered from the most relevant retrieved passages — all computed on-device. Folders are managed on the Knowledge base page, and each answer lists the passages it used.
Session history
Conversations are saved automatically and can be resumed from the sidebar, so nothing is lost between sessions. Search past chats, or delete them all from Settings.
Verifiably offline
An in-app indicator and a hand-maintained dependency allowlist confirm no network-capable plugin is registered in the build. Block your network — Rafeeq runs identically.
Good to know
What Rafeeq does and doesn't do today, stated plainly.
Chat and explain only
Rafeeq never executes code, in code-help mode or anywhere else.
CPU only, no GPU needed
Small quantized GGUF models run on ordinary machines through llama.cpp.
You bring the model
The standard installer includes none. The other installer includes a small starter model, Qwen2.5-0.5B-Instruct.
Long documents are truncated
Only the beginning of a long document fits the model's context (about 2,000 words). Rafeeq tells you when this happens.
Windows first
macOS and Linux builds are on the way.
See it run with the network off.
Install Rafeeq, block network access, and chat.