Add goose and Ollama as first-class onboarding providers

[james-see]

Summary

Goose was intentionally hidden from the desktop onboarding screen — only Claude Code and Codex were listed as visible runtimes (ONBOARDING_RUNTIME_ORDER = ["claude", "codex"]). This excluded the most popular open-source agent harness from the setup flow.

This PR adds goose as a first-class onboarding runtime and adds Ollama as a first-class LLM provider option.

Changes

Goose onboarding visibility (onboardingRuntimeSelection.ts):

• Add "goose" to ONBOARDING_RUNTIME_ORDER so it appears in the harness selection and default-config steps alongside Claude Code and Codex

Ollama LLM provider (agentConfigOptions.tsx):

• Add "ollama" to KNOWN_LLM_PROVIDER_IDS
• Add { id: "ollama", label: "Ollama (local)" } to PERSONA_LLM_PROVIDER_OPTIONS
• Add ollama entry to PROVIDER_CREDENTIAL_CONFIG with no required env keys (Ollama runs locally at localhost:11434 by default, no API key needed)
• No Rust backend changes needed — the goose_requirements readiness function already handles unknown providers via a catch-all _ => {} branch, and goose's file config tier (~/.config/goose/config.yaml) silences provider/model requirements

Tests (onboardingRuntimeSelection.test.mjs):

• Updated visibility, ordering, and readiness tests to reflect goose inclusion

Tests run

• pnpm check (biome + file-sizes + px-text + pubkey-truncation): pass
• node --test onboardingRuntimeSelection.test.mjs (4 tests): pass

Manual validation

Built and ran the desktop app from source (just desktop-standalone) against a local relay:

• Goose appears as "READY" in the harness selection step
• Ollama appears as a provider option in the default model config step
• Goose correctly reads provider/model from ~/.config/goose/config.yaml — no credentials needed for Ollama
• Agent (goose + Ollama) successfully connects to relay, discovers channels, and responds to @mentions via the ACP harness

Docs updated

No docs changes needed — this is a UI visibility/config change, not a feature behavior change.

Risks / assumptions

• Goose was intentionally excluded from onboarding. The original test explicitly asserted runtimeIsVisibleInOnboarding("goose") === false. There may be a product reason for this (e.g. goose onboarding UX is incomplete, or Block wants to promote Claude Code/Codex first). If so, this PR can be scoped to just the Ollama provider addition and goose visibility can be revisited separately.
• Ollama provider has no credential requirements. If the relay or agent harness needs to validate provider credentials for Ollama in the future, the readiness function would need updating.

Plan reference

No plan file — this is a small, well-scoped change (11 lines across 3 files).

[tkhumush]

@james-see I am assuming this is only to put ollama in the short list of providers? right now selecting a custom llm provider and setting it to http://localhost:11434 as a provider does not work.

[james-see]

@james-see I am assuming this is only to put ollama in the short list of providers? right now selecting a custom llm provider and setting it to http://localhost:11434 as a provider does not work.

It is to get all that working yes. So both ollama in the dropdown and then it works confirmed. You can build this branch yourself and try it. It also adds goose as a harness available on the front screen too next to codex and claude. Because i dont understand forcing a 3rd party company involved at all for a supposed open app.

[tkhumush]

Yeah, weird call not add Ollama and Goose from the start.

There is a Goose agent that could be added, but that strangely clunky too.

I’ll try the PR.