Model gateway

Underneath the assistant, GoWork runs a model gateway: one local endpoint that everything — your coding CLIs, API clients, and the assistant itself — can point at. Instead of juggling separate keys and accounts per tool, you get a single, controllable front door for every model.

One endpoint, many protocols

The gateway speaks the major LLM API shapes, so most tools connect with just a base-URL change:

Anthropic Messages
OpenAI Chat Completions
OpenAI Responses
Codex
Gemini

Point Claude Code, Codex, Gemini CLI, or any OpenAI/Anthropic-compatible app at http://localhost:8081, and GoWork translates between whatever the client speaks and whatever the upstream expects.

Account pooling

Add several accounts (for the coding assistants) and API keys (for providers), and the gateway rotates across them — so you aren't bottlenecked on a single account, and tokens are refreshed automatically.

Key management

Provider API keys live on your machine and are managed in one place, rather than copied into every tool's config. Add, remove, and organize them from the desktop app.

Model mapping and routing

Map the model names your tools ask for onto the models you actually want to serve — swap a provider, pin a version, or route a name to a local model — without touching each tool's settings. Local model runtimes (for example Ollama) can be routed too.

Usage and cost tracking

Every request through the gateway is accounted for, so you can see usage and cost across tools and accounts in one place, and keep request logs locally for debugging.

Local and private

The gateway runs at 127.0.0.1:8081. Credentials, keys, and logs stay under ~/.gowork on your machine. Calls go directly from your machine to each provider.

See also: Assistant capabilities · Reference