July 2026 — Pureframe AI 1.0
Pureframe AI is out of beta. Everything below ships as part of 1.0: Official SDKs — the Python SDK (pip install pureframe-ai) and JavaScript SDK (npm install @pureframeai/sdk) are now published, generated directly from the OpenAPI spec so they always match the live API.
MCP server: upload + job tools — the MCP server now includes upload_video (index a local file straight from your AI agent) and get_job_status, alongside search, collections, and video tools. Ask Claude to “upload this folder and tell me when it’s searchable.”
Trial lifecycle — trial accounts now get reminder emails 3 days and 1 day before expiry, plus a 48-hour grace period after — no more surprise hard stops mid-project.
Storage billing — Pay as you go accounts are billed for actual bytes stored at $0.06/GB-month, accrued daily. Deleting videos frees the charge immediately. See Usage and Billing.
Low-balance email alerts — Pay as you go accounts now get an email when their credit balance drops below $1, so you don’t get caught by a mid-project INSUFFICIENT_CREDITS error.
Turnstile bot protection — sign-in and account creation are now protected by invisible Cloudflare Turnstile verification, with no impact on legitimate users.
Improved search recall — search queries now go through LLM-based query expansion before ranking when the literal query finds nothing, merged with the original results — broader, more forgiving matches for short or ambiguous queries.
New onboarding walkthrough — first-time users now see a guided welcome tour highlighting collections, search, and Frame Map.
June 2026
Image search refinements — reverse image search no longer requires acollection_id (search your whole library with an image), and now rejects reference images smaller than 64×64 pixels to avoid low-quality matches.
May 2026
SigLIP visual encoder — Pureframe AI’s visual search moved from CLIP to SigLIP (google/siglip-so400m-patch14-384, 1152-dim embeddings), improving match quality across the board. Existing libraries were re-indexed automatically.
Hybrid search modes — search requests now accept a modes parameter instead of a single mode string, so you can restrict a query to specific signal types or search all of them at once.
Per-segment frame thumbnails — search results now return a thumbnail for each matched segment individually, instead of one thumbnail per video.
Frame Map (Beta) — a new spatial canvas view that projects every indexed frame onto a 2D map organized by visual similarity, with clustering, a minimap, and a preview strip. See Frame Map.
Looking for the full commit-level history? Pureframe AI’s core services are under active development — this page covers customer-visible changes only.