- Home
- Skills
- APIs & Backend
- binquery
binquery
Local CLI for splitting long videos and querying footage folders using OpenCLIP visual intent.
Free
Works with the AI Agents you already use
binquery
Example session with this skill installed
I have a long recording of a city walk in ./raw/walk.mp4. Split it into 10-second clips, index them, and find shots of 'people crossing the street'.
- Read your context and instructions
- Compiled the binquery
I have processed the video file, created the index, and queried for your specific intent. The long recording was split into segments, and the local index has been verified.
Clip Shortlist: 'people crossing the street'
| Path | Score | Gate | Reasons |
|---|---|---|---|
./split-out/walk_005.mp4 | 0.82 | pictorial | High visual match for crowd movement at intersection. |
./split-out/walk_012.mp4 | 0.79 | pictorial | Matches urban crossing environment. |
./split-out/walk_008.mp4 | 0.75 | person_clip | Multiple subjects detected in motion across frame. |
./split-out/walk_021.mp4 | 0.71 | pictorial | Distant view of pedestrian crossing. |
./split-out/walk_003.mp4 | 0.68 | pictorial | Street level view with active foot traffic. |
./split-out/walk_015.mp4 | 0.65 | motion_label | Handheld movement following subjects across street. |
./split-out/walk_009.mp4 | 0.62 | pictorial | Frame includes crosswalk markings and pedestrians. |
./split-out/walk_022.mp4 | 0.59 | pictorial | Near-miss; shows street traffic with some pedestrians. |
Next steps
- Review the clips in
./split-out/to confirm the best takes for your edit. - Run
binquery listto see if any frozen director intents match your next scene. - Use
binquery querywith a different intent sentence to find specific lighting or angles.
Connects securely to your tools. The creator never sees your data.
About this skill
Searching through hours of raw footage for a specific shot or breaking a long recording into manageable chunks is a manual bottleneck for editors. binquery automates the local clip management workflow by providing a CLI-based shortlist generator that ranks footage based on your visual intent.
What it does
- Time-grid splitting breaks one long local video file into segments using ffmpeg.
- Local indexing generates CLIP vectors for a folder of videos without using cloud APIs.
- Intent-based querying ranks 8–15 clips based on a single descriptive sentence.
- Health monitoring verifies index integrity with a doctor command to prevent failed queries.
- Multi-modal scoring provides results with a path, score, gate type, and logic-based reasons.
How it works
- split: Run the command on a long recording to generate a folder of segments based on a time grid.
- index: Process the clip folder to create local vector embeddings and metadata.
- doctor: Verify the generated index files to ensure the query engine has the required data.
- query: Input a visual description to receive a ranked shortlist of local video paths.
Frameworks & tools
This tool requires Python 3.10+, ffmpeg, and ffprobe on the system PATH. It uses OpenCLIP (ViT-B-32) for local embedding generation.
Why this beats prompting it yourself
Writing a custom script to bridge ffmpeg and OpenCLIP requires significant boilerplate for frame extraction and vector management. This skill provides a structured CLI that handles the grid-splitting, local caching, and scoring logic out of the box.
Use cases
- Generate a shortlist of potential reaction shots from a streamer's VOD.
- Find specific visual scenes within a large folder of dailies or rushes.
- Break down a long master file into a grid for faster manual review.
- Organize scene-sequence visual intents during a complex edit.
Known limitations
Does not perform highlight detection, silence cutting, or automatic timeline assembly. Requires local disk space for the index and an initial download for CLIP weights.
Local machine only. No cloud API or auto-editing. Split is time-grid only, not highlight detection. Requires ffmpeg and ffprobe. First run requires internet for weights.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 month ago
- Free with an Agensi account
Requires
Permissions
No special permissions declared or detected