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

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Apr 5, 2026 6:39pm

PR Summary

Medium Risk
Changes API-key requirement logic used for block validation/UI, which can inadvertently hide/show credential fields or fail validations for certain model strings (especially in self-hosted/server contexts). Covered by new unit tests but affects a central model-selection path.

Overview
Fixes server-side/API-key validation for Ollama models in self-hosted Docker deployments by making shouldRequireApiKeyForModel treat models as Ollama when OLLAMA_URL is set and the model is not a known cloud model and has no provider prefix.

Reworks provider detection to prefer the client providers store (case-insensitive) over getProviderFromModel regex inference, and uses getBaseModelProviders() plus the “slash-prefixed = cloud provider” convention to decide when an API key is still required. Adds the isOllamaConfigured feature flag and a comprehensive utils.test.ts suite covering hosted/self-hosted, Azure, vLLM, and Ollama paths.

^{Reviewed by Cursor Bugbot for commit 8225115. Configure here.}

Greptile Summary

This PR fixes a Docker deployment regression where Ollama models (e.g. mistral:latest) incorrectly triggered "Agent is missing required fields: API Key" validation errors. The root cause was getProviderFromModel relying on regex pattern matching, which misidentified Ollama model names as cloud provider models when the Zustand providers store was empty on the server.

Key changes:

• Replaced getCurrentOllamaModels (Ollama-only store slice) with getProviderFromStore (iterates all provider buckets) for client-side resolution
• Added isOllamaConfigured feature flag (Boolean(env.OLLAMA_URL)) for a clean server-side fallback path
• In the isOllamaConfigured server path: slash-prefixed models and models present in getBaseModelProviders() (explicit definitions only, no regex) require an API key; everything else is assumed to be an Ollama model and does not
• 24 new unit tests covering all provider paths, including the regression case of mistral:latest with OLLAMA_URL set

Correctness notes:

• getBaseModelProviders() correctly excludes ollama, vllm, and openrouter from the cloud model map — only explicit model IDs are matched (no regex), preventing the original false-positive
• The slash convention cleanly separates dynamic cloud providers (fireworks, openrouter, groq, cerebras, azure, etc.) from bare Ollama/vLLM model names
• isOllamaConfigured is intentionally server-only; client-side the providers store is hydrated and handles resolution correctly via getProviderFromStore

Confidence Score: 5/5

Safe to merge — the fix is targeted, well-tested, and the logic is sound

All prior P1/P2 concerns from previous review rounds have been addressed and explained by the author. No new P0 or P1 issues found. The three-layer resolution strategy (store → isOllamaConfigured+explicit map → require key) correctly handles all provider paths including the regression case.

No files require special attention

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[shouldRequireApiKeyForModel] --> B{model empty?}
    B -- yes --> C[return false]
    B -- no --> D{isHosted?}
    D -- yes --> E{in hostedModels?}
    E -- yes --> C
    E -- no --> F{vertex/ or bedrock/?}
    D -- no --> F
    F -- yes --> C
    F -- no --> G{isAzureConfigured\nAND azure prefix/model?}
    G -- yes --> C
    G -- no --> H{vllm/ prefix?}
    H -- yes --> C
    H -- no --> I[getProviderFromStore]
    I --> J{storeProvider?}
    J -- ollama or vllm --> C
    J -- other provider --> K[return true]
    J -- null --> L{isOllamaConfigured?}
    L -- no --> K
    L -- yes --> M{has slash?}
    M -- yes --> K
    M -- no --> N{in getBaseModelProviders?}
    N -- yes --> K
    N -- no --> O[return false\nOllama model assumed]

_{Reviews (5): Last reviewed commit: "fix: handle vLLM models in store provide..." | Re-trigger Greptile}

@greptile

@cursor review

@greptile

@cursor review

@greptile

@cursor review

@greptile

@cursor review