Your data never leaves
Agents, knowledge, and conversations stay on your infrastructure. Nothing is shipped to a third-party cloud you do not control — the answer to data residency and on-premise AI requirements.
Your infrastructure, your keys
Self-hosted AI agents — your infrastructure, your keys, your data.
Run AI agents on your own infrastructure with your own LLM keys. One Docker command, any supported LLM, and your data never leaves. Open source, on-premise-ready, no vendor lock-in.
Control, cost, sovereignty
Why self-host AI agents?
Control, cost, and sovereignty — without giving up capability.
Your own LLM keys, no markup
Bring your own keys for OpenAI, Anthropic, Google Gemini, Azure OpenAI, Ollama, OpenRouter, or any OpenAI-compatible provider such as DeepSeek and Mistral. You pay your LLM provider directly — no per-token platform markup, no metered runtime.
You own it — no lock-in
Config-as-code in your repo. Open-source core under BSL 1.1, converting to Apache 2.0. It keeps running regardless of what happens to any vendor.
Built in Go — production-grade
A single compiled binary: fast, concurrent, and lightweight, with per-tool 30-second timeouts, circuit breakers, and LLM retry built in. Not a prototyping toolkit — infrastructure built to run in production.
Read it first
Open-source AI agents you can read before you run
The engine your agents run on is public on GitHub. Read the exact production infrastructure before you trust it — see that it scales, that it is real, that it is not amateur glue. Most platforms are a black box until you are already committed.
BSL 1.1 today, Apache 2.0 after four years. Free to self-host, embed, and modify.
Quick start
Self-hosted AI setup: one Docker command
Two containers — the engine and PostgreSQL. A working agent answering over a REST API in minutes.
Terminalcopy and run
curl -fsSL https://syntheticbrew.ai/releases/docker-compose.yml -o docker-compose.yml && docker compose up -dFull guide in the quick start. Prefer Kubernetes or a standalone binary? Both are supported.
The trade you do not have to make
Self-hosted AI vs closed cloud agent platforms
| Dimension | Self-hosted SyntheticBrew | Closed cloud agent platforms |
|---|---|---|
| Where your data lives | On your infrastructure — the runtime and PostgreSQL run inside your own security perimeter, on-premise or in your private cloud. | On the vendor’s cloud, under their terms, in the regions they offer. |
| Cost model | Free to self-host. You pay your LLM provider directly — no per-token platform markup, no metered runtime. | Metered per request, per token, or per seat, with the platform sitting in your cost path. |
| Inspectability | Open core — read the exact runtime your agents run on before you commit, and modify it. | A black box you cannot audit until you are already committed. |
| Model choice | OpenAI, Anthropic, Google Gemini, Azure OpenAI, Ollama, OpenRouter, or any OpenAI-compatible provider — including fully local models via Ollama. | The models the vendor offers, on the vendor’s schedule. |
| Lock-in | Config-as-code in your repo. BSL 1.1 converts to Apache 2.0 after four years. | Migrating away means rebuilding agents, knowledge, and integrations elsewhere. |
Questions
Self-hosted AI: common questions
What does self-hosting an AI agent mean?
Self-hosting means the agent runtime runs on infrastructure you control — your own servers, your VPC, or on-premise — rather than a third-party cloud. With SyntheticBrew it is a Docker Compose stack (the engine plus PostgreSQL), so your data and your LLM keys never leave your environment.
Can I run AI agents on-premise or air-gapped?
Yes. SyntheticBrew runs anywhere Docker runs — Linux, macOS, or Windows (WSL2) — including on-premise and private networks. Point it at a local model via Ollama for a fully private, air-gapped deployment, or at any hosted LLM with your own keys.
Can I build a self-hosted AI chatbot into my product?
Yes. Your product talks to the runtime over a REST API and receives structured server-sent events, so a self-hosted AI chatbot can live in your own frontend. You can also use the included web client or embed the generated widget.
Is it really open source?
The engine is source-available under BSL 1.1, which converts to Apache 2.0 after four years. You can read the exact production runtime your agents run on before you commit — and modify it.
How much does self-hosting cost?
The Community Edition is free to self-host, with no per-agent, step, session, or storage limits. You only pay your own LLM provider for tokens — there is no per-token platform markup.
Your infrastructure. Your keys. Your agent.
Download the Community Edition and have a self-hosted agent answering over a REST API in minutes — or read the engine first.