fix(llm): pin LLM edge functions to US/EU regions to fix geo-block 403s

[koala73]

Summary

• Root cause: OpenRouter returns 403 "This model is not available in your region" when Vercel routes LLM-calling edge functions through nodes in regions where Google Gemini is geo-blocked (Middle East, East Asia, etc.)
• Affected endpoints: /api/chat-analyst, /api/news/v1/* (article summarization), /api/intelligence/v1/*
• Fix: Pin the three edge functions to ['iad1', 'lhr1', 'fra1', 'sfo1'] (Virginia, London, Frankfurt, San Francisco) where all OpenRouter-routed LLM providers have full availability
• Also: Improves callLlmReasoningStream error logging to include model name + full response body on non-2xx for easier future diagnosis

Diagnosed by pulling Vercel production logs which revealed:

[SummarizeArticle:openrouter] API error: 403 {"error":{"message":"This model is not available in your region.","code":403}}

Test plan

• Deploy and trigger /api/chat-analyst from a VPN endpoint in a geo-blocked region (e.g. UAE, Turkey)
• Confirm no llm_unavailable errors in Vercel function logs
• Confirm article summarization still works globally
• Verify latency impact is minimal (LLM inference dominates, not routing hop)

[vercel]

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

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
worldmonitor	Ignored		Mar 30, 2026 6:52am

[greptile-apps]

Greptile Summary

This PR pins three LLM-calling Vercel edge functions (/api/chat-analyst, /api/news/v1/*, /api/intelligence/v1/*) to US and EU regions (iad1, lhr1, fra1, sfo1) to prevent 403 geo-block errors from OpenRouter when Vercel routes requests through nodes in restricted regions (Middle East, East Asia, etc.). It also improves error logging in callLlmReasoningStream to capture the model name and first 300 characters of the response body on non-2xx responses, making future diagnosis faster.

• Region pinning is applied consistently to all three affected edge functions with the same four-region list.
• Error logging improvement reads the response body via resp.text() only when resp.body is non-null, guarded with .catch(() => '') so any abort or read failure degrades gracefully to an empty string.
• The clearTimeout is called before the new await resp.text() read; the client AbortController signal remains active and can still interrupt a hanging body read if the client disconnects — acceptable for a typically-small error payload.
• No changes to request handling logic or auth; the fix is purely infrastructural.

Confidence Score: 5/5

Safe to merge — targeted infrastructure fix with no logic changes and proper error handling guards.

All four files contain only mechanical, low-risk changes: three one-line config additions and one tightly-scoped logging improvement. No auth, caching, or data-path logic is touched. The .catch(() => '') guard in the logging path prevents any new failure mode. No P0 or P1 findings.

No files require special attention.

Important Files Changed

Filename	Overview
api/chat-analyst.ts	Adds regions: ['iad1', 'lhr1', 'fra1', 'sfo1'] to the Vercel edge config to pin this LLM-calling endpoint away from geo-blocked regions.
api/intelligence/v1/[rpc].ts	Identical region-pinning fix applied to the intelligence RPC gateway edge function.
api/news/v1/[rpc].ts	Identical region-pinning fix applied to the news RPC gateway edge function (article summarization).
server/_shared/llm.ts	Improves non-2xx error logging in callLlmReasoningStream to include the model name and first 300 chars of the response body, aiding future geo-block diagnosis.

Sequence Diagram

sequenceDiagram
    participant Client
    participant Vercel_Edge as Vercel Edge<br/>(iad1/lhr1/fra1/sfo1)
    participant OpenRouter
    participant LLM as LLM Provider<br/>(e.g. Gemini)

    Client->>Vercel_Edge: POST /api/chat-analyst<br/>or /api/news/v1/* or /api/intelligence/v1/*
    Note over Vercel_Edge: regions config pins routing<br/>to US/EU nodes only

    Vercel_Edge->>OpenRouter: POST streaming request<br/>(model, messages, stream:true)
    OpenRouter->>LLM: Forward to LLM provider

    alt Success (2xx + body)
        LLM-->>OpenRouter: SSE stream
        OpenRouter-->>Vercel_Edge: SSE chunks
        Vercel_Edge-->>Client: data: {"delta":"..."}
        Vercel_Edge-->>Client: data: {"done":true}
    else Non-2xx or no body (e.g. 403 geo-block)
        OpenRouter-->>Vercel_Edge: HTTP 4xx/5xx
        Note over Vercel_Edge: clearTimeout, read errBody<br/>log: HTTP 403 model=X body={"error":...}
        Vercel_Edge->>OpenRouter: Retry next provider
    end

Loading

_{Reviews (1): Last reviewed commit: "fix(llm): pin LLM edge functions to US/E..." | Re-trigger Greptile}