Portable context, lower drift.
Cassette.mp4 packages transcripts, summaries, and instructions into a single assistant-readable context envelope.
Architecture/research
A public format concept and roadmap. Portability and reduced drift require validation with actual artifacts.
Availability note: the descriptions and commands below are the existing public concept record. No downloadable cassette release or signed example is supplied here; compatibility and verification remain unverified until tested with an actual artifact.
What Cassette.mp4 is
Cassette.mp4 is a portable, self-describing context envelope for AI assistants. It packages transcripts, summaries, and instructions into a single file that an assistant can consume on drop. The goal is repeatability, low drift, and portability.
- Condenses: long transcripts and videos can be semantically chunked and summarized into compact payloads.
- Faithful: an 80/20 approach keeps the small cassette useful enough for accurate reasoning.
- Portable: travels through email, chat, or uploads without losing structure.
- Universal: encapsulated in an MP4 container for compatibility across platforms.
Why it works
AI drift often comes from feeding models raw, oversized transcripts or unstructured context. Cassette.mp4 reduces that risk with semantic chunking, summarization, and a predictable JSON schema.
This turns hours of dialogue or gigabytes of video into a slimmer, replayable context file.
File anatomy
<ClientName>.mp4 # primary drag-and-drop cassette file
cassette.json # same payload in JSON for editing or audit
schema.json # optional JSON-LD / metadata sidecar
script_tag.txt # optional ready-to-paste script tag
README.txt # one-minute install and validation notesMemory model
- Episodic: references to past interactions or actions.
- Semantic: stable knowledge extracted or linked.
- Procedural: prompts, tools, and guardrails.
- Working: per-task scratchpad for the current objective.
Example uses
- Meeting recording summaries.
- Video condensation into structured JSON.
- Business onboarding packages.
- Knowledge bundles for consistent assistant recall.
Roadmap
- Optional encryption and signed manifests.
- Multi-module bundles with schema, images, and billing.
- Web demo player that renders onboarding UI directly from the cassette.
Developer note
This page intentionally abstracts packaging and extraction specifics to preserve IP while communicating the concept and benefits.