Agent Media Utilities. Three things an autonomous agent needs and could not get anywhere on Pocket's Agentic Portal as of 2026-09-26 (checked against the live 99-service catalog before building this): POST /v1/pdf-render (pdf_base64, page?, dpi?, format?) renders one real page of a PDF to a PNG/JPEG image via poppler. POST /v1/image-convert (image_base64, to, quality?, max_width?) converts between image formats -- png, jpeg, webp, tiff, avif -- including real HEIC/HEVC photos (the format virtually all iPhones produce), which the portal's existing OCR/chart services don't handle. POST /v1/audio-transcribe (audio_base64, language?) submits a short audio clip for transcription and returns a job_id; POST /v1/audio-transcribe-status (job_id) polls it -- split into submit+poll because a real speech-to-text pass cannot reliably finish inside a gateway's relay timeout (design rule 7), so the work runs in the background instead of blocking the relay. Every conversion runs through real tools (poppler, libvips/sharp, libheif, ffmpeg, whisper.cpp) on the request's actual bytes -- nothing here is a stub, a guessed format, or a cached-forever answer.
