Zonic Nandemo · Zonic Dekiru
From raw footage to usable intelligence.
Understand hours of video, find evidence, answer long-form questions, build editable sequences, and produce professional outputs — in the Workbench or through the open Platform API.
Zonic Nandemo
AI's Five Senses
Zonic Dekiru
AI's Brain
How it works
From footage to answers and production
01
Connect Your Footage
Use managed Cloud storage, a Hybrid proxy workflow, or a Local workspace. Zonic registers the sources without pretending every mode moves data the same way.
02
Understand and Reason
Transcripts, scenes, people, metadata, and semantic indexes turn long recordings into searchable evidence, answers, comparisons, and editorial decisions.
03
Review and Produce
Review and revise an editable sequence, render it, or export to a supported editor. Human corrections create the signal for better editorial skills without claiming automatic learning that has not shipped.
Architecture
Two systems: one sees, one thinks
No single model sees everything, and each one fails differently. Zonic splits the job in two — and the second half exists to catch what the first half got wrong.
Zonic Nandemo
AI's Five Senses
Turns unordered raw footage into structured, searchable understanding — every modality read at once, by the model best suited to it.
Visual understanding — scenes, objects, actions
Speech — transcripts, speaker diarization, voiceprints
Faces — detection, identity matching, tracking
Metadata — time, location, camera, topic
Zonic Dekiru
AI's Brain
Reasons across everything Nandemo extracted, cross-references the signals to correct what any single model got wrong, and only then commits an edit.
Cross-modal reasoning — links people, speech and scenes
Correction by triangulation, not by majority vote
Precise retrieval — the right moment, not the right file
Conversational editing — say it, and it cuts
Products and execution
Two products. Three ways to run the work.
Workbench and Platform share intelligence kernels, but they are separate products. Cloud, Hybrid, and Local decide where media and compute live — including browser and user-device capacity that can reduce managed infrastructure cost.
Zonic Workbench
The interactive Next.js product for projects, conversation, human review, sequence revision, preview, rendering, export, and execution-mode choice.
Zonic Platform API
A separately consumable REST, MCP, and connector product for headless ingestion, long-video understanding, retrieval, and supported asynchronous jobs.
Distributed execution modes
HORIZONTAL CLOUD SCALE
Cloud
Managed storage and independent cloud jobs provide horizontal scale. The browser coordinates uploads and can preprocess video to reduce transfer size.
CLIENT-DISTRIBUTED SCALE
Hybrid
Full-resolution media work stays near local files while proxies and scalable intelligence use the cloud. Each user contributes CPU and storage, lowering cloud-heavy media cost.
USER-DEVICE SCALE
Local
Media and production workloads use the user's filesystem, CPU, and storage. It scales to that machine and is not yet fully offline because selected model providers still require network access.
Capabilities
What Zonic can do
Ready capabilities, conditional workflows, and explicit non-capabilities — every status comes from one shared capability contract.
Ready
Projects and Source Libraries
Organize footage and metadata by stable project and source identity in Workbench or through the Platform API.
Ready
Scene Understanding
Detect and describe actions, people, locations, objects, and visual events with precise source time ranges.
Ready
Long-Video Semantic Search
Find visual moments and spoken passages by meaning across long recordings and multiple sources, with timestamped evidence.
Ready
Evidence-Backed VideoQA
Ask questions across scenes, transcripts, and people through Workbench, Project MCP, or the account connector.
Conditional
Cross-Source Reasoning
Compare and synthesize evidence across sources when the required indexes are complete; output quality remains task-dependent.
Ready
Automated Rough Cut
Dekiru turns analyzed footage into an editable sequence using task-specific editorial skills and source evidence.
Ready
Human Sequence Review
Inspect, reorder, trim, and revise an AI-authored sequence while preserving the human correction signal.
Ready
Premiere Pro Export
Generate a relinkable .prproj with supported sequence structure, media, captions, audio, and graphics.
Conditional
CapCut and JianYing Drafts
Install a supported draft when the target application and local filesystem are available; parity is target-dependent.
Conditional
Open Platform API
Run headless ingestion, analysis, retrieval, and supported asynchronous agent jobs as a separate product from Workbench.
Not shipped
Automatic Finished-Project Learning
Automatically learning from arbitrary completed editor projects is part of the vision, but is not shipped today.
From the Team
Latest from Zonic
Tutorials, release notes, and insights from the engineering team.
Vision
The Learning Loop
Human review creates training signals today. Automatic learning from arbitrary finished projects remains the next loop, not a shipped capability.
Expert Edits
Raw footage in
Zonic Sequence
AI rough cut
Editor Refines
In Premiere Pro
Delta Decoded
What changed?
Model Improves
Training signal
Better Sequences
Smarter cuts
Today
Creators edit. Zonic assists.
- ✓
Automated rough cut from raw footage
- ✓
Multi-source merge and scene understanding
- ✓
Premiere Pro .prproj export
- ✓
Trainer UI for explicit feedback
Vision · Not Shipped
Automatic learning from finished projects.
- →
Decode finished .prproj files — no explicit feedback required
- →
Direct supervised learning from professional editors
- →
Every timeline edit is a training signal
- →
Continuously improves across all users
Demo
See Zonic in Action
Ready to get back to the story?
Join creators and studios already using Zonic — or book a call to see how it fits your workflow.