Goose via Tokonomix

//What is Goose?

Goose (by Block) is an open-source autonomous agent that installs, runs and tests code on your machine — CLI and desktop. It uses the OpenAI provider for any OpenAI-compatible endpoint, and its capabilities extend through MCP.

• Runs long autonomous sessions: plans, edits, runs commands and reacts to output.
• Selects a provider via goose configure or environment variables.
• For OpenAI-compatible endpoints, set OPENAI_HOST to the endpoint URL.
• Extends through MCP servers, so a consensus tool fits right in.

//Why route Goose through Tokonomix?

• Goose runs long unattended chains with tool calls — drive it with a direct model like anthropic/claude-sonnet-4-6, and call consensus as an opt-in cross-check on planning-only steps.
• One key, 130+ models — change the model without touching providers.
• Per-call spend caps bound an autonomous run; cost is metered in your dashboard.
• Region / origin routing for a code-executing agent.

//Connect Tokonomix to Goose

Select the OpenAI provider and set OPENAI_HOST to the Tokonomix endpoint. Goose normalises a trailing /v1, so the path resolves correctly to chat completions.

# host + key via environment (NOT config.yaml)
export OPENAI_HOST=https://tokonomix.ai/api/v1
export OPENAI_API_KEY=tok_live_your-key

# ~/.config/goose/config.yaml (or via `goose configure`)
GOOSE_PROVIDER: openai
GOOSE_MODEL: anthropic/claude-sonnet-4-6
# Opt-in (planning/review only, no tool-calling): GOOSE_MODEL: tokonomix-consensus

//Verify it works

• Run goose session and give it a small task.
• Confirm it plans and executes using the Tokonomix model.
• Check the call and token usage in your Tokonomix dashboard.