Introduction

Zonic exposes two API surfaces: Nandemo for video analysis, and Dekiru for editing, QA, and sequence generation on top of analyzed footage.

Two API Surfaces

Zonic integrations usually use one or both product layers. Nandemo turns uploaded footage into structured video understanding. Dekiru uses that understanding to answer questions, search across evidence, and create or revise edit sequences.

Nandemo is the analysis layer. It receives raw video sources and converts them into timestamped records that applications can query directly: scenes, transcript words, speakers, people, metadata, and search indexes. Dekiru is the reasoning layer on top of those records. It does not treat a video as one opaque file; it works from the structured evidence produced by Nandemo.

  • Nandemo: upload footage, run selected analysis jobs, retrieve scenes, transcript words, metadata, labels, people, speakers, and search results.
  • Dekiru: answer VideoQA questions through chat, generate rough-cut sequences, revise plans, upload references, and render an edit. MCP is available for structured retrieval tools.

Core Objects

Project

A project is the top-level namespace. All footage, analysis outputs, transcripts, people, chat history, and sequences live under one pid.

Source

A source is one uploaded video, addressed by its original video name. API paths use the URL-encoded name as {vid}.

Source names are intentionally stable because many outputs are keyed by source name: scene lists, transcript segments, face appearances, and search results all point back to the original footage.

Understanding

Understanding is Nandemo's analyzed, timestamped representation of footage. Build timeline search, clip browsing, review UIs, and agent context from /understanding, video details, transcript words, and people endpoints.

The main unit inside understanding is the scene. For each video source, Zonic separates the timeline into multiple scenes, analyzes each scene, and represents the video as an ordered collection of timestamped observations. This is why downstream features can search for moments, not only whole files.

Which Interface Should You Use?

  • Use GET /projects/{pid}/understanding when your frontend needs raw shot-level data.
  • Use POST /projects/{pid}/videos/search when you want direct semantic search results.
  • Use POST /projects/{pid}/chat when you want Dekiru to answer footage QA/VQA questions or build/revise edit sequences conversationally.
  • Use POST /projects/{pid}/mcp when an MCP client wants structured read-only retrieval tools.

Lifecycle

The API is stateful. A normal integration should call endpoints in this order:

  • Create a project with POST /projects.
  • Register videos with POST /projects/{pid}/videos.
  • Upload video bytes to the returned signed URL.
  • Optionally rerun selected Nandemo jobs with POST /projects/{pid}/videos/{vid}/analysis/jobs.
  • Read Nandemo outputs with GET /projects/{pid}/understanding.
  • Ask Dekiru QA or request sequence generation through POST /projects/{pid}/chat.
  • Use render endpoints after chat has produced a sequence you want to export.

OpenAPI answers field-level questions. These guide pages answer sequencing questions. Keep both open while integrating.