Requirements

Before using this skill, ensure you have:

• sops: brew install sops (for decrypting secrets)
• Wrangler: npm install -g wrangler
• Node.js: brew install node

Must run from the pollinations repo root.

For changes to providers, fallbacks, regions, GPU hosts, or data handling, follow Data-processing documentation maintenance and update the legal register with the operational change.

Architecture Overview

The former enter-services / enter-services-staging EC2 boxes (text-pollinations.service, image-pollinations.service) are decommissioned; text and image generation run inside the gen.pollinations.ai Worker. The Discord bots that lived on enter-services now run on the monitoring-agents EC2 box — see apps/discord-bot-family/README.md and operations/infrastructure/gpu/GPU_INSTANCES.md.

Deploy Cloudflare Worker (enter.pollinations.ai)

Production deploys only through GitHub Actions (Deploy / Cloudflare production, dispatched from the production branch) — never wrangler deploy --env production from a local machine. See AGENTS.md "Cloudflare Production Deployment Safety".

Staging deploys through the Deploy / Cloudflare staging workflow (workflow_dispatch: pick the branch and the service). It runs migrate:staging first, then deploy:staging. Tick push_secrets to also push secrets/staging.vars.json to the Workers; a new or changed value needs Secret Mutation Safety approval (AGENTS.md) before it is merged. The same npm scripts also work locally per service:

cd enter.pollinations.ai
npm run deploy:staging   # staging only

Check what a Worker has with wrangler secret list --env staging. Details: token-rotation.md.

Testing on Staging

• Test token: ENTER_API_TOKEN_STAGING in enter.pollinations.ai/.testingtokens. If auth fails, run npm run check-tokens in enter.pollinations.ai before blaming the deploy; tokens expire without the file changing.
• Agents are called by owner/name model id on staging, same as production.
• Cheap read-only checks: wrangler d1 migrations list DB --remote --env staging, wrangler secret list --env staging, and the dispatch namespace list.

Wrangler Configuration

The wrangler.toml contains environment configs:

Token Rotation

Internal trust-boundary tokens (PLN_ENTER_TOKEN, PLN_GPU_TOKEN, Tinybird tokens, SOPS recipients): see token-rotation.md for the inventory, rotation mechanisms, deploy path, and rollback.

OAuth Client Notes

Dated notes from building third-party OAuth clients against enter. Check current source before relying on them.

• Open WebUI 0.11.x (2026-09): refreshes its token only within 5 minutes of expires_at, never in response to a 401 from a model call. A failed refresh deletes its oauth_session row but keeps the user logged in, so a failed refresh and a revoked API key look the same to the user (401 on the next model call). For self-healing: issue a short expires_in so the refresh fires, and return 200 from the refresh endpoint for every live account.
• apikey table: in shared/db/better-auth.ts the owner field is referenceId, not userId. row.userId compiles in loosely typed code and fails at runtime with a misleading "Unauthorized or invalid session" from createApiKey. Run tsc on new files touching this table before testing live.

Notes

• Production deploys on push to production branch
• Staging deploys via the Deploy / Cloudflare staging workflow (manual dispatch), or per service with npm run deploy:staging
• Always test on staging before merging to production
• enter.pollinations.ai handles auth/billing; generation requests are served by the gen.pollinations.ai Worker