An AI agent that never
leaves your terminal —
or your machine.
Orchestra is a privacy-first, terminal-native coding agent. It plans, calls tools, reads your codebase,
and writes code with the same fluency as cloud tools like Claude Code — running entirely on
ollama, on hardware you own.
Everything a cloud agent does. Nothing leaves the machine.
Nine tools, one loop, zero network dependency on your code or your prompts.
100% local & private
No API keys, no subscriptions. Your code and prompts never leave your machine.
Agentic tool swarm
Reads files, lists directories, runs bash, and edits your codebase autonomously.
Artifact system
Ask for an architectural plan or doc — it's generated as a markdown artifact you can view and re-inject later.
Multi-session support
Isolated chat sessions. Jump between debugging and brainstorming without cross-contaminating context.
Context injection
/add file.py feeds files into context. Prefix any host command with ! to inject its output.
Dynamic theming
Switch between Dracula, Nord, Monokai, and Synthwave on the fly, mid-session.
Prompt library
Save frequently used prompts to ~/.orchestra/prompts/ and load them instantly.
Ambient focus
Toggle background ambient music with Shift+Tab to stay in flow.
Permission gate
Dangerous actions like write_file or run_bash require explicit y/n approval before they execute.
The Autonomous Loop.
Orchestra isn't just an autocomplete. It runs in a continuous feedback loop until the goal is achieved.
(Ask human permission if required)
Why Orchestra?
How Orchestra stacks up against other AI coding assistants in the landscape.
| Feature | Orchestra | Claude Code | Antigravity | Codex |
|---|---|---|---|---|
| Privacy / Execution | 100% Local (Ollama) | Cloud (Anthropic) | Hybrid (Depends) | Cloud (OpenAI) |
| Agentic Loop | Yes (Autonomous) | Yes | Yes | No (Autocomplete) |
| Interface | Rich TUI | CLI | IDE Extension / CLI | IDE Extension |
| Cost | Free (Open Source) | Pay-per-token API | Subscription / API | Subscription |
One prompt. A full plan, executed and verified.
A real Orchestra session: the agent breaks the task down, searches, writes the artifact, and marks it done — no manual steps in between.
Slash commands
Type / inside the TUI for a self-documenting command menu with contextual hints.
| Command | Description |
|---|---|
| /help | Show the help menu |
| /session | Manage isolated chat sessions — new, list, delete, switch |
| /artifact | View and inject AI-generated architectural plans and documents |
| /add <file> | Inject a specific file into the agent's context for the next prompt |
| /prompt <name> | Load a saved prompt from your personal library |
| /model <name> | Switch the active Ollama model |
| /theme <name> | Switch the visual theme — dracula, nord, monokai, synthwave |
| /tasks · /goal | View and manage your persistent autonomous task list |
| /context | View context window usage and a visual map of the conversation |
Running in under five minutes.
Install Python 3.10+ and Ollama, then clone Orchestra and you're ready to go.
Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass
and activate the virtual environment again.
"
Orchestra was born from true human-AI collaboration. By blending the raw power of agentic coding with deliberate engineering intuition, we forged a completely unique value proposition. Version 1 is the culmination of 48 hours of relentless building, testing, and manual curation. But this is just the beginning. The upcoming updates will redefine the landscape. Ultimately, my north star remains unchanged: advancing environmental sustainability, championing human-centered AI, and ensuring uncompromising AI safety.