flowchart TB accTitle: NutsNews local-first AI fallback update accDescr { This update makes the Worker support a full local-first AI flow: } A["NutsNews local-first AI fallback update"] --> B["updates/README_LOCAL_FIRST_AI_FALLBACK_UPDATE.md"] B --> C["Auto-generated placeholder diagram"]

NutsNews local-first AI fallback update

This update makes the Worker support a full local-first AI flow:

Article review tries the home server first through POST /review.
If local review fails, the Worker retries the home server once.
If local review still fails and OpenAI fallback is enabled, the Worker falls back to OpenAI.
Accepted article translations already use the same pattern through POST /translate: local first, retry once, then OpenAI fallback.

Important deployment variables

Use these when generating and deploying Workers:

export ENABLE_LOCAL_AI_SECRET_BINDING=true
export AI_PROVIDER="local"
export LOCAL_AI_URL="https://ai.nutsnews.com"
export LOCAL_AI_MODEL="qwen2.5:3b"
export AI_PROVIDER_FALLBACK_TO_OPENAI=true
export AI_REVIEW_CONCURRENCY=1
export ENABLED_SUMMARY_LANGUAGES="fr,ja,de-CH,de,el"
export SUMMARY_TRANSLATION_LIMIT=12

AI_PROVIDER="local" is the setting that moves article review to the home server first.

AI_PROVIDER_FALLBACK_TO_OPENAI=true keeps OpenAI as the safety net when the home server fails.

AI_REVIEW_CONCURRENCY=1 is recommended for a small home server so multiple shards do not overload Ollama.

Deploy Workers

cd /Users/ramideltoro/WebstormProjects/nutsnews2/worker

export ENABLE_LOCAL_AI_SECRET_BINDING=true
export AI_PROVIDER="local"
export LOCAL_AI_URL="https://ai.nutsnews.com"
export LOCAL_AI_MODEL="qwen2.5:3b"
export AI_PROVIDER_FALLBACK_TO_OPENAI=true
export AI_REVIEW_CONCURRENCY=1
export ENABLED_SUMMARY_LANGUAGES="fr,ja,de-CH,de,el"
export SUMMARY_TRANSLATION_LIMIT=12

npm ci
npx tsc --noEmit
npm run generate:wrangler
npm run deploy:all

Verify generated config

grep -n \
  -e '"AI_PROVIDER"' \
  -e '"LOCAL_AI_URL"' \
  -e '"LOCAL_AI_MODEL"' \
  -e '"AI_PROVIDER_FALLBACK_TO_OPENAI"' \
  -e '"AI_REVIEW_CONCURRENCY"' \
  -e '"ENABLED_SUMMARY_LANGUAGES"' \
  -e '"SUMMARY_TRANSLATION_LIMIT"' \
  generated-wrangler/wrangler.shard1.jsonc

You should see:

"AI_PROVIDER": "local",
"LOCAL_AI_URL": "https://ai.nutsnews.com",
"LOCAL_AI_MODEL": "qwen2.5:3b",
"AI_PROVIDER_FALLBACK_TO_OPENAI": "true",
"AI_REVIEW_CONCURRENCY": "1",
"ENABLED_SUMMARY_LANGUAGES": "fr,ja,de-CH,de,el",
"SUMMARY_TRANSLATION_LIMIT": "12"

Runtime logs to expect

A fully local-first successful article should show:

worker.local_ai.article_reviewed
worker.translation.local.article_summary_translated
worker.translation.local.article_summary_translated

If local review fails, fallback should show:

worker.local_ai.fallback_to_openai
worker.openai.article_reviewed

Test one shard

curl "https://nutsnews-worker-1.nutsnews.workers.dev/?limit=1"

Watch these response fields:

"aiProvider": "local",
"localAiCallCount": 1,
"openAiCallCount": 0,
"articleSummaryTranslationCount": 2

openAiCallCount should stay at 0 when local review works. It may become 1 only if local review fails and fallback is used.

Primary diagram