People
Who owns the work, who influenced it and who is waiting.
Kernal by Andes · transcript X-ray
A conversation contains people, organisations, projects, decisions and commitments. Kernal reconstructs that hidden world as a proposed graph you review, own and carry forward.
Proposed first. Human reviewed. Portable by design.
Transcript fragments
The world behind the words
Notes flatten the work into paragraphs. Kernal keeps the structure: who did what, which project it changed, what remains unresolved and where the statement came from.
Who owns the work, who influenced it and who is waiting.
The companies, teams and institutions surrounding each decision.
What is moving, blocked, proposed or quietly abandoned.
What changed, when it changed and which source supports it.
The promises and next actions that disappear between conversations.
The structure connecting the facts, not another folder of notes.
From conversation to owned record
Start with transcripts or an account export when you want a more complete reconstruction.
Kernal identifies candidate people, organisations, projects, decisions and relationships.
Accept, edit or reject what it proposes. A suggestion does not become your record by itself.
Search the reviewed graph, carry it between AI tools and export it without rebuilding from zero.
Working product, not a memory metaphor
Kernal already runs as a knowledge graph and MCP service. The ChatGPT connector is being prepared; the graph, review controls, retrieval and export model exist independently of it.
Kernal stores entities, relationships, sources, actions, goals, projects and decisions as structured records.
The record preserves where a proposed fact came from so a person can inspect and challenge it.
Hybrid search reaches sources, records, memories, actions and working context through one retrieval surface.
Kernal uses the Model Context Protocol and exportable records so accumulated context can move with you.
Ownership is the product
Kernal separates the record from the AI interface. Review the proposed graph, keep the useful structure and download your record when you leave.
Privacy architecture for the consumer connector is still being completed. Kernal does not yet make a zero-knowledge or "not even Andes can read it" public claim.
Questions buyers ask
No. Kernal works with conversations or exports you explicitly provide. It does not silently reach into other accounts or environments.
No. Extracted people, projects, decisions and relationships are proposals. You can accept, correct or reject them before they become reviewed records.
Yes. Reviewed records can be exported as structured data and exposed through the Model Context Protocol, so the useful context is not tied to one interface.
Use non-sensitive conversations for now. The privacy controls and final terms for the ChatGPT connector are still being completed. Andes will publish the handling model before wider release.
The connector preview is available by invitation. The public ChatGPT listing has been submitted and is under review.
Kernal by Andes
Then decide what is worth keeping.