Personal context,
in perspective.
The people, companies and open systems shaping how AI understands us. An industry atlas for building infrastructure around the person.
Temporary research site · Internal draft · Please do not forward
A real category.
A demanding opportunity.
Personal context already sits at the intersection of memory, identity, agent execution and user control. AI Passport, HCP, Honcho, Unabyss, Inrupt and other efforts overlap with substantial parts of the totem thesis.
Our question is how a person can retain useful continuity across independently operated systems: current understanding, scoped disclosure, traceable corrections and reliable outcomes. That combination has to earn its place through better experiences and measurable results.
This revision retains the original map’s 49 entries and its two analytical views. It adds nine relevant systems, an expert directory, a research graph and an explicit positioning comparison. Public claims and our interpretation are distinguished throughout.
Two views of the same ecosystem.
Look at who governs the relationship, what a system primarily does, how far context travels and who uses it. The positions are qualitative editorial judgments, not benchmark scores. A low position does not mean a system cannot update memory.
Person-governed context across independent systems. The shaded band marks our intended scope; it is a design target, not a measured product capability.
The person is the governing center.
We want identity, events, goals and changing circumstances to remain coherent as a person moves between models, agents, apps and devices. Each system receives the context it is permitted to use; new outcomes and corrections can return to the personal side.
Portability, memory updates, temporal graphs and consent each have precedents. The opportunity we are investigating is their reliable operation as a person-governed service across real product boundaries.
Continuity across touchpoints
Connect events and changing situations across services, beyond a static preference profile or a single agent’s chat history.
Control that travels
Make purpose, audience, provenance and the distinction between knowing and acting explicit at integration boundaries.
A useful return loop
Reconcile outcomes and corrections without letting an agent’s unsupported inference silently become a trusted fact.
| System | Governing center | Context scope | Update / return mechanism | Interoperability | Stage |
|---|
At a comparable model, evidence and compute budget: improve task outcomes, reduce repeated explanation, propagate corrections, limit unwanted disclosure and remain worth the integration and consent effort. A capable harness, memory file, retrieval engine and existing personal-context products are serious baselines.
Connect the layers. Respect their boundaries.
These are functional groupings, not claims that the named systems integrate with totem. PCP remains a working name; compatibility with existing contracts is a research and engineering task.
Companies, products and standards.
The original 49 entries are all retained, with updated English descriptions and evidence links. Nine additions connect the event research to context, execution and personal infrastructure. Open any row for its relevance to totem, documented capabilities and primary sources.
| System / initiative | Audience & authority | Publicly documented capability | Stage & evidence | Source |
|---|
“Reviewed” refers to the linked capability description. It does not independently audit funding, customer counts, security guarantees or actual production performance. Archived figures remain labeled as historical notes.
From people to work to systems.
Solid lines connect authors to their published work or documented project roles. Dashed lines show our analytical connections to related systems and totem’s research questions. Select a node to inspect its public source.
People working on the hard questions.
Field leaders come first, followed by researchers, protocol maintainers and builders grouped by their area of work. These profiles do not imply friendship, endorsement or a commitment to join ipo Lab. Publication affiliations are dated to the work; two archive-backed entries are marked.
What is known. What remains open.
Memory and retrieval, context learning, personal representation and user governance are related research streams. Their results support different claims. We should test the transfer between them.
A shared evaluation agenda
Test cross-agent handoff, changed goals, corrected facts, disclosure scope, conflicting updates and migration. Compare the same underlying models and evidence, then measure task success, stale-context errors, correction delay, integration effort, user burden, latency and cost. Keep additional-data advantages separate from architectural advantages.
What changed after checking the sources.
Follow the work back to its source.
Official papers, project documentation, product descriptions, author pages and a small number of explicitly labeled archive-backed references. Source descriptions establish what is publicly stated; our positioning and suggested questions are analysis.
Scope and method
All 49 entries from the initial landscape remain in the directory. Open Source AI Week, TechCrunch Disrupt and AGNTCon / MCPCon research were used to identify relevant public work and organizations. Event inclusion, an invitation, a conversation and a collaboration are distinct facts; private correspondence and relationship records are not reproduced in this atlas.
Current capability descriptions were checked against primary or participant sources. Archived funding and customer numbers have not been independently re-audited. Coordinates and adjacency judgments are qualitative. The report is focused, not an exhaustive census of every company or expert in the world.