Release history

2026-06-22 — Inline video frames

Added: video_frames content part for chat completions — pass pre-decoded video frames inline (each with an image_url and a timestamp_ms) as an alternative to a single video_url. Useful when you’ve already sampled frames client-side or need precise control over which frames the model sees.

Added: /v1/files endpoints (upload, list, retrieve, retrieve content, delete) for uploading images and video once and referencing them by id across requests.
Added: Files developer guide and OpenAI-compatible API reference pages.

Added: /v1/detect endpoint for direct grounded object detection — return boxes or points from text categories, annotated exemplars, or exhaustive detection, without building a chat-completions prompt.
Added: Detect capability guide and API reference pages with full request/response schemas.

Added: perceptron-mk1 — flagship VLM with image + video input, 32K context, and reasoning support.
Added: Video capability guides (Video Q&A, Video Clipping, ICL Video) plus matching cookbook notebooks.
Added: Models tab with per-model spec pages (modality, context, pricing).
Updated: Image capability guides rewritten for Perceptron Mk1 and the new SDK API surface.

Added: Guide covering Perceptron MCP setup, Perceptron MCP tools, and tasks best suited for Isaac-powered agents.

Added: isaac-0.2-2b-preview and isaac-0.2-1b image VLMs (2B and 1B parameter, respectively).
Added: Guides covering thinking and tool calling flows.

Added: isaac-0.1 image VLM (2B parameter).
Added: Perceptron SDK quickstart with first-run Python & CLI snippets.
Added: Guides covering captioning, VQA, detection, OCR, ICL, streaming, and edge deployment.
Added: Tutorials for manufacturing defects, OCR & Layout Extraction, and PDF document intake.
Added: Best practices playbooks for prompting, latency optimization, and security.
Added: API reference navigation with endpoint examples and CLI commands.