PyTorchGPU 0 · 8 GB kernel ready
In [1]:
import torch
torch.cuda.is_available()
Out[1]: True
01
Jupyter, on demand.
Spin up a notebook in seconds — PyTorch, TensorFlow, R, or bare Python. Persistent volumes, GPU passthrough, a file browser built in.
Open-source·Self-hosted
Your machine.
Your models.
Your swarm.
Orqestra turns the computer you already own into a full AI platform — notebooks, model hosting, sandboxes, and multi-agent swarms. No rented GPUs. No vendor lock-in.
agenthive / content-swarm
liveCEOResearcher— delegate: gather sources on X
ResearcherWriter— handoff: 6 findings attached
WriterQA— send_message: draft v2 ready
QACEO— return: score 0.94 — ship it
Next.js · tRPC · Go orchestrators · PostgreSQL · Redis · Python agent harness
The platform
One platform. Four ways to run AI.
Each one isolated, resource-capped, and yours.
POST/v1/chat/completions
llama3.1:8bready
mistral-smallready
qwen2.5-coderpulling 71%
02
Host any model.
Pull from a unified catalog, get an OpenAI-compatible endpoint. Your models, your API, running on your silicon.
alpine · sandbox-7f3a
$ ssh root@sandbox-7f3a
identity: ~/.orqestra/keys/7f3a.pem
Welcome to Alpine Linux 3.20
sandbox:~#
03
Disposable Linux sandboxes.
SSH into a fresh container, generated keys, gone when you're done. Host a custom server or just experiment.
04
AgentHive.
Wire specialist agents into a swarm. Watch them delegate, hand off, and collaborate — live.
AgentHive
Stop prompting one model. Orchestrate a team.
Define agents with their own roles, instructions, and models. Connect them with explicit communication flows. Run OpenAI, Anthropic, Gemini, Groq — or your own local models — side by side in one swarm.
swarm / research-pod4 agents · LiteLLM
CEOorchestratoropenai/gpt-4o
Researcherworkeranthropic/claude
Developerworkerlocal/qwen2.5-coder
QAworkergroq/llama-3.1
+ define agent — role, instructions, model
Handoff
Control transfers fully to a specialist. Sequential workflows, tight feedback.
Orchestrator–Worker
One agent delegates, collects, and synthesizes. Parallel, high-autonomy tasks.
Every message streamed in real time. Every step traceable.
Getting started
Running in three steps.
01
Step 1 — Install.
One script provisions Docker, the network, every service.
02
Step 2 — Open the dashboard.
Notebooks, models, sandboxes, swarms — one control plane.
03
Step 3 — Orchestrate.
Build it, run it, watch it stream.
Under the hood
Built like infrastructure, because it is.
Lightweight Go orchestrators drive the Docker socket. tRPC end to end. Postgres for state, Redis for realtime. No magic — just a clean spine you can read.
Web
Next.js 15 · Tailwind v4
API
tRPC v11 · type-safe end to end
Orchestrators
Go · Docker SDK
Data
PostgreSQL 17 · Redis 7
AI tooling shouldn’t mean renting
someone else’s computer.
Orqestra is open-source and self-hosted. Clone it, run it, own it.
bash
$
curl -fsSL orqestra.xyz/install.sh | bash