tl;dr
you write a prompt. an AI agent executes it. code gets written, committed, and deployed. but here's the real thing: you can invite your team, customize your execution environment, manage dozens of agents, set custom rules, and every single action is logged.
live: orquesta.live
npm: npm install -g orquesta-agent
what is orquesta?
orquesta is an open-source "prompt to production" platform. you describe what you want in plain english, and an AI agent executes it — not in a chat window, but on your actual machine or cloud VM. it writes files, runs commands, commits to git, and deploys.
but the real power is what's around the agent: a full control plane for AI-assisted development.
why i built this
i was tired of the copy-paste loop with AI chatbots. but the bigger problem was this:
i had a team. i had multiple projects. i had VMs running in the cloud. and i wanted AI agents working on all of them — but i needed control.
who can execute prompts? what environment are they running in? what did the agent actually do? can i set rules before anything ships? can i let a client or contractor use my infrastructure without giving them SSH access?
nothing out there solved this. so i built orquesta.
let others work in your environment
this is the core idea. you set up a machine — your laptop, a cloud VM, a staging server — and install the orquesta agent. now anyone you invite can submit prompts that execute on that machine, without needing direct access.
how it works
- you install the agent:
npm install -g orquesta-agent && orquesta-agent --token YOUR_TOKEN - the agent connects to your orquesta dashboard
- you invite team members, contractors, or clients to your project
- they submit prompts from the dashboard (or telegram, or the CLI)
- the agent executes on your machine. they never touch your server directly.
this means:
- contractors can work on your codebase without SSH keys
- clients can request changes without knowing how to code
- team members can ship features from their phone via telegram
- you keep full control over what runs and where
customize your execution environment
every project is different. orquesta lets you configure exactly how the agent behaves.
- CLAUDE.md sync — define project rules, coding standards, and context in a CLAUDE.md file. it syncs automatically to the agent so every prompt execution follows your conventions
- environment variables — set env vars from the dashboard. the agent picks them up without you SSHing into the machine
-
execution modes — choose how prompts run:
- auto — AI analyzes the task and picks the best method
- SSH — direct single command execution
- agent — full Claude CLI with file access, git, everything
- batuta — autonomous loop that keeps working until the task is done (max 15 iterations)
- cloud VM provisioning — spin up DigitalOcean droplets directly from the dashboard, pre-configured with the agent and your settings
your environment, your rules. the agent just follows them.
manage dozens of agents
this is where it gets powerful. you're not limited to one agent on one machine.
- agent grid — see all your connected agents in a real-time dashboard. drag, resize, rearrange. each cell shows live terminal output
- multi-project — each project gets its own agent, its own config, its own team
- status monitoring — see which agents are online, idle, or executing. reconnection and heartbeats are handled automatically
- interactive sessions — open a live terminal to any agent. work alongside the AI in real time
imagine having 10 projects, each with an agent running on a different VM, and you can see all of them from one screen. submit a prompt to any of them. watch them all work.
keep all the control
AI agents executing code on your infrastructure? you need guardrails. orquesta gives you full control.
quality gates
before a prompt executes, the AI simulates the changes and shows a preview. team leads can review and sign off before anything touches the codebase. no yolo deploys.
role-based permissions
not everyone should have the same access. assign roles — admin, developer, viewer. control who can submit prompts, who can review, and who can deploy.
custom rules via CLAUDE.md
define what the agent can and can't do. coding standards, forbidden patterns, required tests, deployment rules — all in a single file that the agent respects on every execution.
everything is logged
every. single. thing.
- prompt history — who submitted what, when, and why
- execution logs — full terminal output, streamed in real time and stored permanently
- git commits — every change the agent makes gets committed with a proper message. full diffs available
- token usage & costs — track how many tokens each prompt consumed and what it cost
- activity feed — a chronological feed of everything that happened across all your projects
- performance dashboard — execution times, success rates, cost trends over time
full audit trail. if something goes wrong, you can trace exactly what happened, who triggered it, and what the agent did step by step.
more features
- real-time streaming — watch the agent work live. every keystroke, every file change
- telegram bot — submit prompts and get updates from your phone
- embeddable SDK — drop orquesta into your own app with a few lines of code
-
CLI tool —
orquesta-clifor local development with LLM config sync - batuta autonomous mode — AI loops until the task is done, showing reasoning at every step
the stack
- frontend: Next.js, React, Tailwind CSS
- backend: Next.js API routes + WebSocket server
- database: Supabase (postgres + realtime)
- ai: Claude CLI (via Anthropic) + OpenAI for routing
- agent: Node.js package, runs anywhere
try it out
orquesta is open source. self-host it or use the hosted version at orquesta.live.
if you're managing multiple projects, working with a team, or just want more control over how AI agents work on your code — give it a try.
feedback and ideas are welcome. this is a community project.
built with late nights, too much coffee, and the belief that AI agents should work for you — not the other way around.
Top comments (0)