# Orquesta

> Prompt to Production: a B2B platform where developers and their teams write prompts that drive AI agents to build, review, and ship software. Agents run on customer machines, prompts and logs are tracked in a multi-tenant dashboard, and the LLM layer is multi-model and multi-tier (Claude, GPT-4o, Gemini, DeepSeek, plus local models via Ollama / vLLM).

Orquesta has three surfaces:

- The **dashboard** at getorquesta.com (Next.js + self-hosted Supabase) — runs prompts, browses logs, manages projects and team members.
- The **orquesta-cli** (npm: `orquesta-cli`) — a local TUI coding assistant with a multi-role orchestration engine (planner + parallel workers + refiner) that can use any OpenAI-compatible model.
- The **Batuta LLM** proxy — an OpenAI-compatible `/v1/chat/completions` endpoint with smart per-prompt classification across model tiers (fast / balanced / premium). Used by both the CLI and any third-party app.

## Public API

- [OpenAPI spec (JSON)](https://getorquesta.com/openapi.json): machine-readable spec for /api/v1/*.
- [OpenAPI spec (YAML)](https://getorquesta.com/openapi.yaml): human-readable equivalent.
- [API reference](https://getorquesta.com/docs/api): rendered ReDoc with examples for every endpoint.

The public API covers: `POST /api/v1/chat/completions` (Batuta LLM proxy), `GET /api/v1/models`, `GET /api/v1/prompts`, `GET /api/v1/prompts/{id}`, `GET /api/v1/logs`, `GET /api/v1/projects`, `GET /api/v1/agents`, `GET /api/v1/users`.

Auth: `Authorization: Bearer <token>` or `X-API-Key: <token>` with token prefixes `oak_` (org API key), `oclt_` (CLI), or `oat_` (project-scoped agent).

## MCP Server

Orquesta speaks the Model Context Protocol at `https://getorquesta.com/api/mcp` (Streamable HTTP, stateless). Any MCP client — Claude Code, Cursor, VS Code, Lovable, ChatGPT — can connect and read a project's history or write work back into it. Auth is OAuth 2.1 (dynamic client registration + PKCE, so a hosted client connects from the URL alone) or any bearer token above.

Tools: `list_projects`, `recent_context`, `list_prompts`, `get_prompt_logs`, `submit_prompt` (queues work for the project's local agent), `report_session` (records work done elsewhere, never executed).

- [MCP docs](https://getorquesta.com/docs#mcp): connecting a client, tool reference, scopes.

## Docs

- [Getting started](https://getorquesta.com/docs#getting-started): install, sign up, first prompt.
- [Local Agent](https://getorquesta.com/docs#agent): install `orquesta-agent`, connect it to a project, execute prompts.
- [Orquesta CLI](https://getorquesta.com/docs#cli): install, configure LLM endpoints, sync with the dashboard.
- [Batuta LLM](https://getorquesta.com/docs#batuta-llm): smart routing across model tiers, multi-role orchestration (planner / executor / refiner).
- [API Keys](https://getorquesta.com/docs#api-keys): generating, scoping, and using the three token types.
- [Team Collaboration](https://getorquesta.com/docs#collab): roles, project chat, prompt comments.

## Install

- CLI: `npm install -g orquesta-cli`
- Agent: `npm install -g orquesta-cli orquesta-agent` then `orquesta-agent --token <oat_token>` (orquesta-cli provides the `orquesta` command, orquesta-agent the daemon)
- Web: sign up at https://getorquesta.com/signup

## Optional

- [PromptOps](https://getorquesta.com/promptops): what PromptOps means and how Orquesta implements it.
- [Open-source feed](https://getorquesta.com/oss): community-published prompts and projects.
- [GitHub (OSS edition)](https://github.com/Orquesta-live/orquesta-oss): self-hosted alternative built on Next.js + Prisma + Better Auth.
- [Pricing](https://getorquesta.com/#pricing): Starter (free), Pro ($6/user/mo with 500K tokens of Batuta chat), Max ($12/user/mo — Batuta Standard 5M tokens included + Orquesta CLI + cloud VMs). Add-ons: Batuta Plus ($49, 15M), Pro ($99, 40M), Team ($199, 100M).
- [Contact](mailto:oscar@getorquesta.com): enterprise sales, custom integrations, security review.