Inspect the source.
Inspect duration, resolution, audio, captions, and local sidecars before deciding what the task actually needs.
python skill/scripts/voidscape.py inspect "clip.mp4"
Make the media you keep legible.
Voidscape turns a local file or URL into grounded frames and timestamped text—while keeping cost, cloud processing, and model downloads visible before work begins.
local-firstaudio stays yours by default
cost-awarepreview before processing
agent-nativeframes + citations, not guesswork
Three deliberate moves. One clear decision.
Start with source facts, preview the boundary, then prepare artifacts an agent can inspect.
Preview the tradeoff.
Separate transcription spend from vision-token cost, flag dependencies, and show whether approval is required.
python skill/scripts/voidscape.py preview "clip.mp4" --tier both --backend captions
Read with consent.
Extract only the frames and transcript you approved, then hand agents artifacts they can cite and inspect.
python skill/scripts/voidscape.py read "clip.mp4" --tier both --backend captions
See the cost before you press play.
Preview is a decision point, not a receipt. It exposes the frame plan, transcription path, cloud boundary, and model-download state before reading changes anything.
$ voidscape.py preview "clip.mp4" --tier both
source scanduration + captions
frame plancount + vision tokens
transcription pathlocal / cloud / download
consent staterequired if applicable
decisionvisible before read
Nothing is impliedGATEDAn API key in the environment is not consent. A cloud path or first model download must be approved explicitly.
What can read now—and where the system can grow.
Every direction is explicitly labeled: working today, designed next, or longer-term exploration.
Read the media you already have.
A local-first pipeline for evidence, pricing, and agent-ready outputs.
Local video and recordings
Analyze demos, meetings, tutorials, screen captures, and personal archives from a local file.
Public video URLs
Bring a supported video URL into the same inspect → preview → read path, including individual YouTube links.
Audio, captions, and transcripts
Extract audio artifacts, use available captions, or transcribe locally before cloud analysis is approved.
Instagram URL helper
A repository-only helper validates and deduplicates confirmed public Reel URLs; it is not an installed Voidscape command.
Evidence an agent can use
Give an agent timestamped frames, transcripts, and manifests instead of asking it to guess what happened in a video.
Add trusted ways in.
The next steps extend source capture without changing the consent-first reading core.
Private YouTube queue capture
Queue videos from a user-owned private playlist through the official Data API, then keep the normal analysis flow.
Image and carousel reading
Apply the same inspect-before-send model to visual posts, slides, and image collections.
More audio source adapters
Bring saved talks, podcasts, and voice material into a consistent local transcription workflow.
Reusable capture adapters
Extract a shared interface only after the Instagram and YouTube paths prove the common behavior.
A broader personal media system.
Directions worth testing after product, privacy, platform, and legal review.
More saved-content sources
Potential adapters for X bookmarks, TikTok favorites, LinkedIn saves, newsletters, and RSS collections.
Posts, articles, and threads
Extend the evidence model beyond media to web reading, written research, and saved knowledge.
Agent tools and MCP packaging
Make approved analysis artifacts easier for agent environments to discover and use.
Guided installation and hosted product
Lower the setup barrier only after the local product, user controls, and operating model are ready.
Universal extension and multi-model reading
Research directions—not current commitments—until their safety, cost, and platform constraints are validated.
Built from footage, not hypotheticals.
Voidscape is useful when a recording contains information an agent cannot safely infer from a title, thumbnail, or prompt.
Saved learning video to research note.
Bring a confirmed public Reel or video URL into a local queue, inspect it through the normal gate, then keep a timestamped note instead of another forgotten save.
Voice memo to project decision.
Transcribe a recorded brainstorm locally, preserve the evidence, and let an agent compare ideas against the words that were actually said.
From clone to first proof.
Install Voidscape, optionally set your local Library, then use a key-free fixture to inspect, preview, and read evidence. No paid or remote step is implied.
Install Voidscape for Codex and shared agents.
The installer adds the guided Voidscape skill plus the legacy compatibility entry point.
$ .\scripts\install-skill.ps1
Choose an Inbox and Library when you need them.
Review local folders and transcription defaults first. Preferences save only with explicit confirmation.
$ python skill/scripts/voidscape.py customize
Use original footage, then read the evidence.
The fixture needs no account, key, or copyrighted media.
$ python scripts/create-demo-fixture.py
$ python skill/scripts/voidscape.py inspect samples/build-week-demo.mp4
$ python skill/scripts/voidscape.py preview samples/build-week-demo.mp4 --tier both --backend captions
$ python skill/scripts/voidscape.py read samples/build-week-demo.mp4 --tier both --backend captions --workdir samples/build-week-output
What the agent reads
Open
manifest.json, transcript.txt, and frames/, then answer with [MM:SS] citations. It reads evidence instead of guessing what happened in a video.
Questions worth answering before you hand media to an agent.
Voidscape makes the boundary visible: what stays local, what costs, and what exists today.
Does Voidscape upload my media?
Not by default. Local files, sidecar transcripts, and local transcription stay on your machine. Any cloud transcription route needs a current --allow-cloud approval; a first local model download needs its own --allow-model-download.
What files does “read” create?
A work folder containing selected frames, a transcript when requested, and a manifest that maps frames to timestamps. An agent can turn those artifacts into a Markdown note; the CLI does not pretend to write a semantic note itself.
When should I inspect, preview, or read?
Inspect is free source discovery. Preview is the cost and privacy decision. Read only follows once the path is acceptable and prepares the evidence an agent needs.
Can I use it for Instagram research or voice notes?
Audio-only reads use the installed CLI. Instagram support is a repository-only URL helper and user-observed development workflow, not an installed command.
What if the media needs my login?
Public URLs are the baseline. For media you are permitted to access, the CLI can optionally use a local cookie export while a browser-connected agent works from tabs only after your site approval. Voidscape never extracts your browser credentials, and first-party account connections are future work.
What about Substack, RSS, scheduling, or a browser extension?
Those are planned directions. They are not part of this installed release, and text articles are not sent through the video engine.