Skip to content

Latest commit

 

History

247 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Swobu — LLM Switchboard

English · 简体中文 · 日本語 · Português (Brasil) · Bahasa Indonesia · 한국어 · Русский · Español · Українська

Pool LLM capacity you already have.

Put provider accounts, cloud regions, hosted endpoints, and local GPUs behind stable model names. Your agents keep one endpoint; Swobu handles routing, runtime fallback, and protocol translation underneath.

Documentation · Quickstart · VS Code · Releases

Sharing an HTTPS endpoint, switching its backend without changing the remote client, and revoking access with Swobu

Claude Code, Codex, Antigravity and other clients route through Swobu to provider capacity including OpenAI, Anthropic, Gemini, Bedrock, Azure AI, Mistral, DeepSeek and Ollama

Compatibility shown, not partnership. Support varies by provider and runtime.


Install

macOS, Linux, or WSL:

curl -fsSL https://swobu.com/install.sh | sh

Windows PowerShell:

irm https://swobu.com/install.ps1 | iex

Cockpit opens after install. Add a provider, create a route, then connect an agent.

swobu connect claude
swobu connect codex

Muse, pi, Kilo, and Hermes are also supported. Certified with Antigravity CLI 1.2.3: run swobu connect antigravity --workspace <name>, open a new terminal, then run ordinary agy.

Build your first route in five minutes →


Route once. Change the capacity underneath.

A Swobu route looks like a model name to your agent. It is the stable name the agent already sends while Swobu owns the configured capacity behind it.

agent: model = coding
          │
          ▼
        Swobu
          │
          ├─ Azure AI / region A
          ├─ Azure AI / region B
          ├─ AWS Bedrock
          └─ Ollama

Routes can pool provider accounts, regions, hosted endpoints, local servers, and models. Peer target order changes across attempts; fallback tiers define what happens after a real attempt fails. Swobu does not inspect live quota, price, health, or latency.

Runtime fallback, not preflight. Swobu does not preflight compatibility. It sends the real request to the configured target, and a failed attempt can advance to the next target in the route.

Protocol translation at one boundary. Current protocol families include OpenAI Responses, OpenAI Chat Completions, Anthropic Messages, and Gemini Interactions. Translation occurs where the requested semantics are representable.

How routing works → · Capability matrix →


Share a route without sharing provider credentials

swobu share dev/coding

A Share gives the remote client an HTTPS endpoint and bearer. Provider keys remain on the machine running Swobu. You can change what backs the route without changing the remote configuration, then revoke access when you are done.

Application TLS terminates on the owner machine running Swobu; the relay forwards encrypted application traffic and does not terminate that TLS.

Workspace and Route Share →


Local-first

Swobu runs locally and exposes the endpoint your agents use. Provider credentials stay at that boundary. Requests leave your machine when a hosted target is selected. No Swobu account is required for local use. Operational telemetry is deliberately limited and can be disabled.

Security & privacy →


Releases

Swobu publishes versioned Linux, macOS, and Windows binaries with SHA-256 checksums.

Latest release →

Build from source:

git clone https://github.com/swobuforge/swobu.git
cd swobu
make build
./.out/swobu --version

About

LLM switchboard for Claude Code, Codex and AI agents — pool providers, cloud credits, regions and local models behind stable routes.

Topics

Resources

Security policy

Stars

18 stars

Watchers

0 watching

Forks

Releases

Contributors

Languages