Swobu

Use the AI coding client you like with the backend you want.

Swobu is a local router for AI coding clients. Point Claude Code, Codex CLI, Continue, or another OpenAI/Anthropic-compatible client at Swobu, then route traffic to OpenAI, Anthropic, OpenRouter, AWS Bedrock, Ollama, ChatGPT, or a custom backend.

The client is not the brain. Swobu is early. Expect sharp edges.

---

Install

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

The installer uses GitHub release artifacts (checksums.txt verified).

If a release artifact is not available yet, install from source:

go install github.com/swobuforge/swobu/cmd/swobu@master
swobu

Check health:

swobu status

Stop Swobu:

swobu down

---

60-second setup

Run Swobu:

swobu

In the cockpit:

1. Create or select a workspace.
2. Choose a backend provider.
3. Configure base URL and auth.
4. Copy the local endpoint shown by Swobu.
5. Point your AI client at that endpoint.

Typical OpenAI-style client config:

OPENAI_BASE_URL=http://127.0.0.1:7926/v1
OPENAI_API_KEY=swobu

Anthropic-style clients use the corresponding Swobu-supported Anthropic surface.

Exact environment variables depend on the client.

---

Demo

---

Why Swobu

AI coding clients are becoming frontends.

But each client still tends to carry assumptions about provider choice, API shape, streaming behavior, auth handling, model names, and error shape.

That breaks down when you want to:

• keep your preferred client
• switch model backends
• route local and hosted models through one boundary
• compare backends from the same client experience
• inspect local traffic health
• avoid rewriting workflow glue every week

Swobu standardizes the boundary, not the tool.

---

What works today

Swobu is beta. These surfaces are the current focus.

Tested clients

• Claude Code
• Codex CLI
• Continue
• OpenAI-compatible clients
• Anthropic-compatible clients

Supported backends

• OpenAI
• Anthropic
• OpenRouter
• AWS Bedrock
• Ollama
• ChatGPT
• Custom OpenAI-compatible backends

Supported request families

OpenAI-style:

• /v1/chat/completions
• /v1/responses
• /v1/completions

Anthropic-style:

• /v1/messages

Streaming

• Server-Sent Events
• WebSocket

Known gaps

• compatibility varies by client and backend
• some clients require client-specific environment variables
• provider-reported token/cache fields are not uniform
• release installer support is currently Linux/macOS (amd64/arm64)

---

Cockpit

Swobu includes a local terminal cockpit for setup and operation.

Use it to:

• create, rename, and delete workspaces
• choose backend/provider per workspace
• configure routing and auth references
• inspect readiness and health
• inspect traffic outcomes
• open help and feedback actions

Cockpit is the primary local operator surface.

---

Command surface

swobu                         # launch cockpit
swobu daemon                  # run daemon in foreground
swobu status                  # inspect daemon health
swobu down                    # stop daemon
swobu telemetry status        # inspect telemetry setting
swobu telemetry on            # enable telemetry
swobu telemetry off           # disable telemetry
swobu version                 # print version

Help:

swobu --help
swobu daemon --help
swobu status --help

---

Scripted / non-interactive use

Run the daemon directly:

swobu daemon

Use an explicit config path:

swobu daemon --config /path/to/swobu.yaml

Check from scripts or CI:

swobu status

Health semantics:

• exit 0: healthy
• exit 1: running but uninitialized or degraded
• exit 2: daemon not reachable

Shutdown:

swobu down

---

Run from source

Latest master:

go run github.com/swobuforge/swobu/cmd/swobu@master --help

Install from source:

go install github.com/swobuforge/swobu/cmd/swobu@master
swobu --help

Use this when you want source behavior instead of the install-script channel.

---

What Swobu is

Swobu is:

• a local AI compatibility layer
• a protocol shim
• a client/backend boundary
• a local operator cockpit
• a way to hot-swap LLM backends behind existing AI clients

What Swobu is not

Swobu is not currently:

• an SDK
• a hosted model marketplace
• a new AI client
• an observability platform
• a prompt management system
• a managed enterprise gateway

---

Security and privacy

Swobu is local-first.

By default, Swobu:

• binds to loopback
• keeps control on your machine
• does not send prompts, completions, or auth material through default telemetry

Telemetry can be turned off:

swobu telemetry off

Telemetry details:

• docs/telemetry.md

Local-first is not offline-only.

If you configure a hosted backend, AI requests still go to that backend.

---

Roadmap direction

Near-term focus:

• deeper client profiles
• deeper backend profiles
• better config generation
• better compatibility diagnostics
• clearer error translation
• stronger streaming support
• safer local defaults
• easier backend hot-swapping

Goal:

Make it boring to connect any supported AI client to the backend you choose.

---

Contributing

Contributions are welcome.

Swobu uses a Contributor License Agreement.

By submitting a pull request or other contribution, you agree to the terms in CLA.md. This allows Swobu to maintain, sublicense, dual-license, and relicense contributions in the future.

Why a CLA?

Swobu is AGPL-licensed, and may also offer commercial licensing for teams that cannot use AGPL software. The CLA preserves that option while keeping the public repo open. Contributors keep ownership of their contributions.

Read CONTRIBUTING.md before opening a pull request.

---

Security

Do not report security vulnerabilities in public issues.

See SECURITY.md.

---

Commercial licensing

For commercial licensing and additional permissions:

• contact@swobu.com

---

License

Swobu is released under AGPL-3.0-only.

See LICENSE.