# PlatPhorm Podcasts agent instructions

## When to use

Use this service to discover, search, normalize, index, and play public, source-attributed podcasts, radio, music, and compatible media feeds. Public GET routes and MCP introspection are read-only.

The primary contract is ONE INPUT, ONE PLAYER, ONE MEDIA GRAPH, ONE CONTEXT MODEL, ONE RANKING ENGINE, and ONE BACKEND TRUTH. Send one natural-language request or public URL to POST /api/ai/chat. It searches canonical records first, performs bounded provider discovery when useful, persists normalized relationships, and returns truthful capability for the persistent player. Resolve, ingest, feed, and search endpoints remain supported compatibility and machine surfaces; do not model them as separate consumer workflows.

Vercel is the experience plane for rendering, public routes, SEO, the universal input, and the persistent player. Cloudflare is the intelligence plane for D1 canonical records, KV derived state, R2 bounded artifacts, Queues, Workflows, search support, provider health, ranking, repair, enrichment, and signed relay.

## When not to use

Avoid attempts to bypass publisher access controls, licensing, geographic restrictions, attribution, or trusted-domain protections. Destructive operations are prohibited on public surfaces.

## Authentication

Public URL submissions, grounded analysis and recommendations, discovery, and playback metadata are intentionally available without a user key. Deletion, bulk insertion/import, administrative actions, and raw backend writes require PLATPHORM_API_KEY through either Authorization: Bearer or X-PlatPhorm-API-Key. Never include credentials in URLs, logs, traces, prompts, or public output.

## Safety and limits

Remote requests, redirects, payload sizes, feed traversal, and queue work are bounded. Localhost, private, link-local, and metadata-network targets are blocked. Media is streamed on demand from attributed sources and is not permanently mirrored unless policy and explicit operator intent permit it.

Start with /llms.txt, /llms-full.txt, /llms-index.json, /openapi.yaml, /.well-known/agents.json, /.well-known/mcp.json, and /api/docs.

Vercel AI Gateway is optional and must never be treated as a source of record. Cloudflare D1 is canonical backend truth; degraded or unavailable dependencies must be reported honestly.