Starport

Starport is a self-hosted LLM inference gateway. It serves OpenAI-compatible routes at /v1 and OpenRouter-compatible routes at /api/v1. One gateway routes requests across every provider in its catalog and picks routes by price, latency, or an explicit order.

Keep the two credential kinds separate. A gateway API key authenticates a client to Starport. A provider credential pays a provider. Never use one where the other belongs.

Install

Install the released cask on macOS or Linux:

brew install agentstation/tap/starport
starport --version

GitHub Releases at agentstation/starport carries checksummed archives for macOS, Linux, and Windows.

Start a development gateway

Export one provider credential, then start an isolated gateway:

export OPENAI_API_KEY="replace-with-provider-inference-key"
starport dev --no-open

The command starts a loopback gateway on port 8080 with in-memory state. It prints one temporary gateway API key and a one-time console launch link. Add --no-open to print the link instead of opening a browser, which fits an agent session. The gateway blocks the terminal, so run it in a background process and keep it running.

Wait for readiness before the first request:

curl --fail http://127.0.0.1:8080/health/ready

Start a durable gateway

For state that survives restarts, initialize once and then serve:

starport init
starport serve

starport init writes local configuration, creates encrypted storage, and prints the first gateway API key once. Store that key: no later command prints it again. When a browser must reach the console without a launch link, it presents this machine's local admin token instead. starport auth token --copy puts that token on the clipboard.

Point a client at the gateway

Use the printed gateway API key as the bearer token.

• OpenAI SDK or harness: base URL http://127.0.0.1:8080/v1.
• OpenRouter SDK or harness: base URL http://127.0.0.1:8080/api/v1.

export STARPORT_API_KEY="replace-with-gateway-api-key"
curl --fail-with-body http://127.0.0.1:8080/v1/chat/completions \
  -H "Authorization: Bearer $STARPORT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"openai/gpt-4o-mini","messages":[{"role":"user","content":"ping"}]}'

Model IDs use the provider/model form. A bare model name is not routable.

Answer catalog questions

The CLI answers catalog questions offline from the embedded catalog. Use --json for machine-readable output.

starport models search gpt-4o --json
starport models show openai/gpt-4o-mini --json

models search matches every query word against model IDs, names, and authors. It answers a compact summary: ID, name, context length, and token prices. models show answers the full projection for one exact model ID. That projection carries prices, capabilities, modalities, and every routable provider offering.

Diagnose a deployment

starport doctor --json
starport config validate
starport config show --json

starport doctor runs read-only checks on paths, configuration, the master key, the catalog, and provider adapters. Add --probe to inspect configured storage and API keys in read-only mode.

Read more

• Run starport agent setup after a CLI upgrade to reinstall this skill.
• starport --help lists every command.
• The repository README at agentstation/starport documents the quick start.
• docs/OPERATOR-GUIDE.md in that repository documents operations: budgets, presets, caching, guardrails, webhooks, and telemetry export.