fix(providers): stop reporting an absent Ollama as an error

[waleedlatif1]

Summary

OllamaModelsAPI is the single largest error stream in the app: 10,068 in 14 days, and every one is the same line — Unable to connect to http://localhost:11434. OLLAMA_URL is unset in production, so the route falls back to a hardcoded loopback default and dials it on every poll.

Its siblings already handle this correctly, and produce zero errors over the same window:

Route	Absent-service handling	Errors / 14d
vLLM	logger.info('VLLM_BASE_URL not configured')	0
LiteLLM	logger.info('LITELLM_BASE_URL not configured')	0
Ollama	none — always dials the default	10,068

Changes

• An unreachable Ollama logs at info, not error. It is optional; a deployment that runs none refuses the connection on every poll, which is an expected state, not a failure of this route. This is the half that helps self-hosters
• The hosted platform skips the probe, since it has no local runtime and the loopback default cannot answer — but only when OLLAMA_URL is unset. An explicit URL states intent and is still honoured, even on hosted
• isOllamaUrlConfigured() sits beside getOllamaUrl() so OLLAMA_URL keeps a single owner, rather than the route reaching around it to read the same variable

The response is { models: [] } on every non-success path, exactly as before — the same value a blacklisted provider already returns.

Nothing changes for self-hosters

All four permutations are exercised as tests:

Deployment	Probes Ollama?	vs. before
Self-hosted, localhost default, no configuration	yes	identical
Self-hosted, explicit OLLAMA_URL	yes	identical
Hosted, no configuration	skipped	the fix
Hosted, explicit OLLAMA_URL	yes	honoured

Verified preconditions rather than assumed them:

• OLLAMA_URL is declared in the env schema (z.string().url().optional()), so the configured-check reads a real value
• isHosted is forceHosted || hostname === 'sim.ai' || *.sim.ai, and forceHosted requires !isProd — a self-hosted production deployment can never satisfy it
• providers/utils already imports env-flags, so the new import adds no dependency weight
• The Ollama provider (providers/ollama/index.ts, the execution path) is untouched; this only affects the model-listing route

Type of Change

• Bug fix

Testing

5 tests. Both halves verified red on their own revert — removing the gate fails the hosted case, restoring logger.error fails the unreachable case. The first version of the level test passed either way (it only asserted the response body), so it now asserts the level itself.

app/api/providers 50 passing / 5 files; wider sweep 1,593 passing. Typecheck, biome, and check:api-validation clean. One unrelated suite fails in the worktree from a tailwind v4/v3 resolution artifact — it fails with my changes stashed and passes on the main checkout (13/13).

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Aug 7, 2026 7:45pm

[cursor]

PR Summary

Low Risk
Logging and optional early-return for the Ollama model-list route only; API responses stay { models: [] } on failure paths and self-hosted probe behavior is preserved.

Overview
Stops the Ollama models API from flooding production logs when no Ollama is present. The route used to always hit the loopback default and log errors on every poll; vLLM/LiteLLM already treat missing optional backends as info and return { models: [] }.

Hosted: When isHosted and OLLAMA_URL is unset, the route skips the fetch entirely (loopback cannot work there). An explicit OLLAMA_URL still probes.

Self-hosted: Behavior unchanged for successful probes; connection failures now log at info (Ollama service is not reachable) instead of error, while bad 2xx bodies still log error (Ollama returned a response this route cannot read).

Adds isOllamaUrlConfigured() next to getOllamaUrl() in urls.ts, and a new route.test.ts covering hosted skip, explicit URL on hosted, self-hosted default, unreachable vs malformed responses, blacklist, and non-2xx.

^{Reviewed by Cursor Bugbot for commit ee8aa7f. Configure here.}

[greptile-apps]

Greptile Summary

This follow-up cleanly separates an absent Ollama service from malformed successful responses while avoiding unnecessary hosted-platform probes.

• Skips the default loopback probe on hosted deployments unless OLLAMA_URL is explicitly configured.
• Logs connection failures at info level while preserving warning-level non-2xx and error-level unreadable-response diagnostics.
• Adds coverage for hosted and self-hosted configurations, connection failures, invalid responses, and blacklist behavior.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
apps/sim/app/api/providers/ollama/models/route.ts	Splits connection, HTTP-status, and response-decoding outcomes so only an absent optional Ollama service is treated as informational.
apps/sim/app/api/providers/ollama/models/route.test.ts	Adds focused tests for deployment modes, explicit configuration, logging levels, response validation, and provider blacklisting.
apps/sim/lib/core/utils/urls.ts	Adds a centralized helper distinguishing an explicit Ollama URL from the loopback fallback.

_{Reviews (2): Last reviewed commit: "fix(providers): keep an unreadable Ollam..." | Re-trigger Greptile}

[waleedlatif1]

@greptile

[waleedlatif1]

@cursor review

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit ee8aa7f. Configure here.}