Use Goose with the Gateway¶

Goose is an open-source AI agent that automates coding tasks from your terminal or desktop. Connect it to the Gateway to run Goose through your regional endpoint with full data residency and budget controls.

Prerequisites¶

Goose installed (installation guide)
An LLM API key and endpoint from my.amazee.io

Option 1 — Use the built-in OpenAI provider¶

Goose's built-in OpenAI provider supports custom endpoints, so you can point it directly at the Gateway.

Goose Desktop¶

Open Settings → Models → Configure providers
Find OpenAI and click Configure
Fill in:

Field	Value
OpenAI API Key	your LLM API key
OpenAI Host	`https://llm.[your-region].amazee.ai`
OpenAI Base Path	`v1/chat/completions` (default)

Save, then pick a model from the list (for example claude-4-5-sonnet)

Goose CLI¶

Run the configuration command:

goose configure

Select Configure Providers, then OpenAI. On a first run, Goose shows a setup wizard instead: choose Manual Configuration, then Search all providers..., type OpenAI and select it.
Answer Yes to Would you like to set OPENAI_API_KEY?, then paste your LLM API key
When asked Would you like to configure advanced settings?, answer Yes
Leave OPENAI_BASE_URL empty
Set OPENAI_HOST to https://llm.[your-region].amazee.ai
Keep OPENAI_BASE_PATH as v1/chat/completions
Press Enter to skip OPENAI_ORGANIZATION and OPENAI_PROJECT. When asked Would you like to set OPENAI_CUSTOM_HEADERS?, answer No (it defaults to Yes, and an empty value isn't accepted). Press Enter to keep the default OPENAI_TIMEOUT.
Goose loads the model list from the Gateway. Type claude-4-5-sonnet to filter the list, then select it.

Answer Yes to advanced settings

The advanced settings prompt defaults to No. The host setting is behind it. If you skip it, Goose keeps the default host and sends your key to api.openai.com.

Replace [your-region] with your region code (e.g. ch103, us103).

Option 2 — Add a custom provider¶

Custom providers let you give the connection a friendly name (e.g. "Gateway") and store a list of preferred models. This is useful if you switch between multiple endpoints.

Goose CLI¶

goose configure

Select Custom Providers, then Add A Custom Provider
Answer the prompts:

Prompt	Value
What type of API is this?	OpenAI Compatible
What should we call this provider?	`amazee.ai` (or any display name)
Provider API URL	`https://llm.[your-region].amazee.ai/v1`
Does this provider require authentication?	Yes
How should goose obtain credentials?	Static API key
API key	your LLM API key
Available models	`claude-4-5-sonnet, claude-4-5-haiku`
Supports streaming?	Yes
API base path	leave empty

Config file¶

You can also create the provider config file directly. Save this as ~/.config/goose/custom_providers/amazeeai.json (macOS/Linux) or %APPDATA%\Block\goose\config\custom_providers\amazeeai.json (Windows):

{
  "name": "amazeeai",
  "engine": "openai",
  "display_name": "amazee.ai",
  "description": "amazee.ai regional AI gateway",
  "api_key_env": "AMAZEEAI_API_KEY",
  "base_url": "https://llm.[your-region].amazee.ai/v1",
  "models": [
    { "name": "claude-4-5-sonnet", "context_limit": 200000 },
    { "name": "claude-4-5-haiku", "context_limit": 200000 },
    { "name": "claude-4-5-opus", "context_limit": 200000 }
  ]
}

Each entry in models must be an object with a name. If you use plain strings, Goose skips the whole file without telling you.

Then set your API key as an environment variable in your shell profile (~/.zshrc, ~/.bashrc):

export AMAZEEAI_API_KEY="your-api-key-here"

Then run Goose with the provider:

goose run --provider amazeeai --model claude-4-5-sonnet -t "list the files in this directory"

Verify¶

Start a Goose session and give it a simple task:

goose

goose> list the files in this directory

Check the My usage card on the AI keys page to confirm requests are flowing through your regional endpoint.

Troubleshooting¶

Auth errors: Verify your API key is correct. For Option 1, check that OPENAI_API_KEY is set. For Option 2, check the environment variable named in api_key_env is exported.
Unknown provider: amazeeai (mentions the keychain): Goose couldn't parse your custom provider file. The keychain is not the problem. Look in ~/.local/state/goose/logs/ for Failed to parse. Usually models is a list of strings instead of { "name": ... } objects, or name or engine is missing.
Model not found or 403 team_model_access_denied: The model ID doesn't exist or isn't enabled for your key. Run curl https://llm.[your-region].amazee.ai/v1/models -H "Authorization: Bearer $AMAZEEAI_API_KEY" and copy an ID from the list.
Connection errors: For Option 1, check that OPENAI_HOST is set to your Gateway URL and not api.openai.com. Make sure the host URL includes your region code and uses https://. For the custom provider config, the base_url should end with /v1.
Tool calling not working: Goose relies on tool calling support. Use Claude models (Sonnet, Haiku, Opus) which have full tool calling support through the Gateway.