fix: harden Ollama autodiscovery and warning behavior by vincentkoc · Pull Request #29201 · openclaw/openclaw

vincentkoc · 2026-02-27T23:10:21Z

Summary

Hardens Ollama implicit provider autodiscovery so local models are registered without requiring OLLAMA_API_KEY.
Keeps implicit probing quiet for users without explicit Ollama configuration, while preserving warnings when explicit config is provided and discovery fails.
Fixes test env restoration in the autodiscovery test so process.env.VITEST is correctly restored/deleted and does not leak as the string "undefined".

Credit

Credit to @marcodelpin for the core approach and implementation direction in fix: Auto-discover Ollama models without requiring API key #20872.

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

User-visible / Behavior Changes

Ollama models can be auto-discovered without pre-setting OLLAMA_API_KEY.
Implicit local discovery remains quiet when Ollama is not explicitly configured.

Security Impact (required)

New permissions/capabilities? (No)
Secrets/tokens handling changed? (No)
New/changed network calls? (No)
Command/tool execution surface changed? (No)
Data access scope changed? (No)

Repro + Verification

Environment

OS: macOS
Runtime/container: Node 22
Model/provider: Ollama
Integration/channel (if any): N/A
Relevant config (redacted): N/A

Steps

Run provider resolution without OLLAMA_API_KEY and with discoverable Ollama models.
Verify providers.ollama is registered with discovered models.
Verify unreachable implicit probe does not warn unless Ollama is explicitly configured.

Expected

Provider is auto-registered when models are discoverable.
No warning noise for non-explicit configs.
Warnings preserved for explicit config failures.

Actual

Matches expected.

Evidence

Failing test/log before + passing after
Trace/log snippets
Screenshot/recording
Perf numbers (if relevant)

Human Verification (required)

Verified scenarios: targeted Ollama provider tests and new autodiscovery tests.
Edge cases checked: explicit config failure warning path; env restoration safety in tests.
What you did not verify: manual local Ollama pull/run on this machine.

Compatibility / Migration

Backward compatible? (Yes)
Config/env changes? (No)
Migration needed? (No)

Failure Recovery (if this breaks)

How to disable/revert this change quickly: revert this PR or set explicit provider config.
Files/config to restore: src/agents/models-config.providers.ts
Known bad symptoms reviewers should watch for: unexpected Ollama warning noise or missing Ollama provider entries.

Risks and Mitigations

Risk: Implicit discovery could still be noisy in edge paths.
- Mitigation: Explicit quiet gating + focused tests for warning behavior.

greptile-apps · 2026-02-27T23:16:40Z

Greptile Summary

This PR successfully hardens Ollama autodiscovery by enabling local model registration without requiring OLLAMA_API_KEY, while keeping implicit probing quiet unless Ollama is explicitly configured.

Key improvements:

Auto-registers Ollama provider when models are discovered locally (no API key required)
Suppresses warnings for implicit local probing when Ollama isn't explicitly configured
Preserves warnings when explicit Ollama config is provided but discovery fails
Fixes test environment restoration to properly handle undefined env vars (prevents leaking "undefined" strings)

Implementation details:

Added quiet parameter to discoverOllamaModels and buildOllamaProvider to control warning behavior
Modified provider registration logic: now registers if models are discovered OR API key is present (previously only with API key)
Uses "ollama-local" as sentinel API key value for auto-discovered providers
Comprehensive test coverage for all three scenarios: successful discovery, quiet implicit probing, and explicit config warnings

The changes are backward compatible, well-tested, and align with the stated objectives. No issues found.