Summary

Fixes #4544
Ref #4782

Ollama is a local service that doesn't require API authentication, but resolveImplicitProviders() only registered the Ollama provider when OLLAMA_API_KEY was explicitly configured. This caused locally installed models to show as "missing" in openclaw models list and fail with "Unknown model" at runtime.

Root Cause

In src/agents/models-config.providers.ts, the Ollama provider registration was gated behind:

const ollamaKey = resolveEnvApiKeyVarName("ollama") ?? resolveApiKeyFromProfiles(...);
if (ollamaKey) { // <-- never true for default Ollama installs
  providers.ollama = ...;
}

Since Ollama runs locally without authentication, most users never set OLLAMA_API_KEY, so discovery never ran.

Changes

• Auto-discover Ollama models by always calling buildOllamaProvider() — if models are found via /api/tags, the provider is registered with a placeholder "ollama-local" API key
• Silent auto-discovery: added a quiet option to discoverOllamaModels() to suppress console.warn when the user hasn't explicitly configured Ollama. This prevents noisy warnings for users who don't have Ollama installed (addresses review feedback on fix: Auto-discover Ollama models without requiring explicit API key #4782)
• Backward compatible: explicit OLLAMA_API_KEY or auth profile configuration still works as before, with warnings preserved when Ollama is unreachable
• 3 new tests validating auto-discovery, silent failure, and explicit-key warning behavior

Files Changed (2)

• src/agents/models-config.providers.ts — provider resolution logic + quiet flag (+28/-10)
• src/agents/models-config.providers.ollama-autodiscovery.test.ts — 3 new tests

Test Plan

• 3 new auto-discovery tests pass (mock fetch, cover all paths)
• 10 existing models-config.providers tests pass
• 28 models list command tests pass
• Manual: install Ollama, pull model, verify openclaw models list shows it without API key

Greptile Summary

Enables Ollama models to be auto-discovered without requiring OLLAMA_API_KEY, fixing a key usability issue where locally installed Ollama models appeared as "missing". The implementation adds a quiet parameter to suppress warnings when Ollama isn't explicitly configured, preventing noise for users without Ollama while preserving warnings for users who explicitly set up Ollama but can't reach it.

• Auto-discovery now runs unconditionally, registering the provider with placeholder "ollama-local" API key when models are found
• Silent mode (quiet: true) when no API key configured prevents unnecessary warnings for users without Ollama
• Backward compatible: explicit API key configuration preserves original warning behavior
• Well-tested with 3 new tests covering auto-discovery, silent failure, and explicit-key warning scenarios

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• The changes are well-isolated to Ollama provider discovery logic, backward compatible, and thoroughly tested. The implementation follows existing patterns (similar to vLLM provider), adds proper error handling with the quiet flag, and includes comprehensive test coverage for all code paths. No security concerns or breaking changes identified.
• No files require special attention

_{Last reviewed commit: 81c8562}

🤖 Generated with Claude Code
Testing: fully tested (automated + manual verification)
I understand and can explain all changes in this PR.