Skip to main content

​
Overview

ondoki can convert video recordings into step-by-step workflows. Upload a video file, and an async processing pipeline extracts frames, transcribes audio, and generates annotated steps with AI.

​
Supported Formats

FormatExtension
MP4.mp4
MOV.mov
AVI.avi
MKV.mkv
WebM.webm
M4V.m4v
Maximum file size: 2 GB

​
How It Works

1

Upload Video

Upload a video file via the Video Import page or the API. The file is validated and saved to disk.
2

Queue Processing

A MediaProcessingJob is created and queued for the Celery media worker.
3

Extract Audio

Audio is extracted from the video for transcription.
4

Transcribe

Audio is transcribed using Whisper (configurable model size: base, small, medium, large).
5

Extract Key Frames

Key frames are extracted from the video at regular intervals.
6

Analyze & Generate

AI analyzes the frames and transcript to generate annotated steps with titles and descriptions.
7

Create Workflow

A new ProcessRecordingSession is created with source_type “video”, containing the generated steps and guide.

​
Processing Stages

You can track progress via the status endpoint:
StageDescription
uploadingFile is being uploaded
queuedJob is queued for processing
extracting_audioAudio track being extracted
transcribingAudio being transcribed to text
extracting_framesKey frames being extracted
analyzingAI analyzing content
generatingGenerating steps and guide
doneProcessing complete
failedProcessing failed (check error)
Progress is tracked as a percentage (0–100%).

​
API

​
Upload

POST /api/v1/video-import/upload
Content-Type: multipart/form-data

file: <video file>
project_id: <project_id>
Returns: session_id, job_id, task_id

​
Check Status

GET /api/v1/video-import/status/{session_id}
Returns: 
{
  "status": "processing",
  "processing_stage": "transcribing",
  "progress": 45,
  "video_filename": "demo.mp4",
  "video_size_bytes": 52428800,
  "estimated_frames": 120,
  "frame_count": 0,
  "has_guide": false
}

​
Check Job Status

GET /api/v1/video-import/jobs/{job_id}

​
Requirements

Video import requires:
  1. Celery media worker — must be running (media-worker service in docker-compose)
  2. Whisper model — configured via WHISPER_MODEL env var (default: base)
  3. OpenAI API key — if using OpenAI’s Whisper API (set OPENAI_API_KEY)
Without the media worker running, video uploads will be accepted but processing will not start. The job will remain in queued status.

​
Whisper Model Sizes

ModelSizeSpeedAccuracy
base74 MBFastGood for clear audio
small244 MBMediumBetter accuracy
medium769 MBSlowerHigh accuracy
large1.5 GBSlowestBest accuracy
Set via WHISPER_MODEL environment variable. Larger models require more RAM and processing time.

​
Retry Behavior

Failed processing jobs are retried up to 3 times (configurable via max_attempts). The job status tracks:
  • Number of attempts
  • Error messages
  • Start and finish timestamps