Features
Everything you need,
nothing you don't.
orangu is deliberately minimal. Every feature earns its place: no bloat, no cloud dependencies, no telemetry.
Streaming AI, on your terms.
OpenAI-compatible
Works with any server that speaks the OpenAI API: llama.cpp, Ollama, LM Studio, and hosted endpoints. Switch servers and models without restarting.
Live streaming
Tokens stream live to your terminal. For llama.cpp, the footer shows native throughput: Working @ 42.5 t/s (2s).
Fully offline
After the initial model download, no internet connection is required. Your conversations, your code, your machine. Zero telemetry.
Persistent sessions
Every session is saved and auto-resumed when you return to the same workspace and branch. Full context is preserved across restarts.
Workspace tools
The model reads, edits, lists, and runs, all within your workspace and within configured limits. You control exactly what it can touch.
Request cancellation
Press Esc Esc to cancel any in-flight request. Queue up the next command while the model is still working.
Git, from a single prompt.
Branch management
List, create, rename, switch, and delete branches. /branch, /checkout, /merge, all with Tab completion.
Commit workflow
/commit, /amend, /squash, /push: the full commit-to-push loop without leaving orangu.
Review mode
Full-screen two-pane diff review of your branch against main. Mark lines, add comments, and request AI feedback. No browser needed.
GitHub and GitLab
Create PRs, check them out by number, close issues, add comments, all wired to gh or glab when installed.
Auto-sync
orangu fast-forwards your default branch against origin in the background on every start, keeping your repo in sync silently.
Cherry-pick and stash
Pick commits from other branches, or stash and restore uncommitted changes, all with Tab completion over commit hashes and stash entries.
A terminal that thinks with you.
Ghost text hints
As you type, a grey inline preview shows the command you're building toward. Tab accepts it; Shift+Tab cycles through alternatives.
Natural language
Type show status or create pull request and they work exactly like slash commands.
Markdown rendering
AI responses render with bold, italic, headings, lists, and syntax-highlighted code blocks, right in the terminal. No browser tab needed.
Per-session history
Command history is saved per session, not globally. Arrow keys browse your session's commands without mixing in other workspaces.
Horizontal panning
Long lines extend beyond terminal width onto a virtual canvas. Pan with Alt+Left/Right without reflowing.
Built in Rust
Fast startup, zero GC pauses, small binary. Tested on Fedora; runs on any Linux. Windows via WSL.
See it in your terminal.
Build from source, run the wizard, and you'll have a full AI assistant in your repo in under five minutes.