PIPELINE-AI-DEPLOYMENT¶

Archived record

This page describes the Firebase-era platform or a migration step that has completed. It is kept as history and is not a current runbook. The current platform is described from the home page.

Last Updated: February 23, 2026 Status: Active (Pre-deploy + post-deploy runbook)

Executive Summary¶

This runbook covers end-to-end AI deployment for ContentForge: configuration, key requirements, readiness validation, and post-deployment verification.

What was requested earlier and now implemented¶

Provider-agnostic model routing (Vertex/OpenAI/Anthropic) — implemented.
Gemini minimum baseline 2.5+ with cost-first defaults — implemented.
Scrape fallback chain (direct → reader → optional headless) — implemented foundation.
Deployment/readiness surface for AI stack — implemented via endpoint + local script.

Remaining non-code dependency from prior requests¶

Headless fallback service itself (Cloud Run/Playwright endpoint) still needs deployment if you want headless_fallback mode active in production.

Required Keys / APIs¶

Required for default path¶

Google Cloud project with Vertex AI enabled
Firebase Functions deployment permissions
REPLICATE_API_KEY (for publish-time hero covers, preview image slots, and curated article images)
A deployed/runtime environment that can resolve Vertex ADC credentials

Required only if provider switched¶

OPENAI_API_KEY when CONTENTFORGE_AI_PROVIDER=openai
ANTHROPIC_API_KEY when CONTENTFORGE_AI_PROVIDER=anthropic

Required for Firecrawl-based ingestion¶

FIRECRAWL_API_KEY

Required only if headless fallback enabled¶

CONTENTFORGE_HEADLESS_FALLBACK_ENABLED=true
CONTENTFORGE_HEADLESS_FALLBACK_URL=https://<your-cloud-run-service>/scrape

Recommended Runtime Configuration¶

CONTENTFORGE_AI_PROVIDER=vertex
CONTENTFORGE_VERTEX_LOCATION=us-central1
CONTENTFORGE_VERTEX_DRAFT_MODEL=gemini-2.5-flash-lite
CONTENTFORGE_VERTEX_ANALYSIS_MODEL=gemini-2.5-flash
CONTENTFORGE_VERTEX_MULTIMODAL_MODEL=gemini-2.5-flash
CONTENTFORGE_IMAGE_MODEL=google/imagen-4-fast
CONTENTFORGE_IMAGE_MODEL_HERO=google/imagen-4-fast
CONTENTFORGE_METADATA_ONLY=true
CONTENTFORGE_LOG_TOKEN_USAGE=true
CONTENTFORGE_ALT_TEXT_ENABLED=false
CONTENTFORGE_IMAGE_FALLBACK_PROVIDER=none
CONTENTFORGE_SCRAPE_FALLBACK_ENABLED=true
CONTENTFORGE_HEADLESS_FALLBACK_ENABLED=false

Local runtime file¶

Create functions/.env for non-secret runtime configuration. Firebase Functions v2 will read this for local emulator/deploy workflows, and standalone local validation can use:

cd functions
node --env-file=.env check-ai-stack-readiness.js

Do not store secrets in functions/.env. Keep API keys in Firebase Functions secrets.

Pre-Deploy Validation¶

Local env snapshot¶

cd functions
npm run readiness:local

Optional remote readiness (after deploy)¶

cd functions
set CONTENTFORGE_ADMIN_BEARER=<firebase-id-token>
npm run readiness:remote

The remote helper derives the endpoint from the Firebase project and region. To target a different deployment, override the defaults:

cd functions
node check-ai-stack-readiness.js --remote --project <project-id> --region <region> --token "<firebase-id-token>"

Browser-backed remote readiness (auto token)¶

If you already have a valid admin browser session, the repo can restore that session and mint a fresh Firebase ID token automatically:

npm run readiness:remote:auto

If this is the first run on the smoke profile, capture storage state once first:

npm run smoke:auth:capture
npm run readiness:remote:auto

From functions/, the same helper is available as:

npm run readiness:remote:auto

Exit code 2 means configuration is incomplete.

Deploy¶

cd functions
npm run deploy

Post-Deploy Verification¶

Trigger and validate draft generation endpoint.
Trigger and validate preview image generation endpoint.
Check aiStackReadiness endpoint as admin.
Run scrape benchmark script against representative URLs:

cd functions
node test-scrape-fallback.js --endpoint "https://<region>-<project>.cloudfunctions.net/generateArticleDraft" --url "https://example.com/article-1" --url "https://example.com/article-2"

Confirm scrape mode distribution (direct_html, reader_fallback, headless_fallback).

Secret setup¶

cd functions
firebase functions:secrets:set REPLICATE_API_KEY --project hybridcloudworks-61e8d
firebase functions:secrets:set FIRECRAWL_API_KEY --project hybridcloudworks-61e8d

Optional only if used:

firebase functions:secrets:set OPENAI_API_KEY --project hybridcloudworks-61e8d
firebase functions:secrets:set ANTHROPIC_API_KEY --project hybridcloudworks-61e8d

Readiness expectations¶

vertex.ready should be true in the readiness output.
REPLICATE_API_KEY is required for the default article pipeline because image generation remains Replicate-backed even when text generation stays on Vertex.
controls.metadataOnly=true is the recommended restart posture to reduce ingest-time spend.

Persona-Informed Art of Possible¶

AAI: Add quality-gated automatic model escalation (flash-lite → flash → pro) based on confidence score and budget envelopes.
CKAD: Deploy headless fallback as autoscaled Cloud Run service with strict timeout/cost caps.
GHE: Add CI gate to fail PR when readiness script returns missing required keys for target env.
KCS: Emit weekly AI ops report from readiness + scrape benchmark outputs.