A brain without a soul
That is the phrase I keep using for this project.
I do not want to build another agent framework that decides how an agent thinks, plans, talks, or orchestrates a little company of other agents.
I want the layer underneath that.
Memory. Context. Ingestion. Durable knowledge. The boring infrastructure that lets an agent stop waking up as a brilliant amnesiac every time a process ends.
Where the idea came from
I have become increasingly dependent on AI agents for real work, especially coding and research. The models themselves keep improving at a ridiculous pace, but the surrounding problem remains familiar: context is expensive, project knowledge is scattered, and useful things an agent learned yesterday often need to be reconstructed tomorrow.
Existing hosted memory products can solve some of that, but cost, control, portability, and long-term trust start mattering once the memory becomes important rather than convenient.
So I started thinking about what I would want underneath whichever agent system I happen to use next year.
The answer is intentionally less glamorous than an autonomous-agent demo.
What belongs here
The platform is meant to provide things such as:
- durable ingestion from sources like repositories, feeds, and websites;
- organized knowledge contexts or “banks”;
- long-lived agent memories that are separate from any one model provider;
- queued jobs for slower background accumulation and maintenance;
- ways to preserve both what went right and what went wrong;
- retrieval and context surfaces that other agents can use without adopting an orchestration framework.
I am interested in agents gradually becoming better informed about a project rather than repeatedly receiving an enormous context dump and being expected to rediscover the same world.
The product problem behind the technical one
The more intimate and durable agent memory becomes, the less comfortable I am with vague promises about ownership and offboarding.
If somebody trusts a system with years of accumulated context, “account terminated, good luck” is not an acceptable product boundary.
So I have spent nearly as much time thinking about export, deletion, encryption, offboarding, billing disputes, backups, evidence of policy versions, and what happens when a user must be removed as I have about retrieval.
This may be the least glamorous paragraph on the page.
It is also probably one of the reasons I still care about the project.
Current direction
Keep the foundation narrow.
Provider APIs above it. Agents above it. Whatever orchestration people prefer above that.
The platform should remember and expose useful context without insisting on becoming the personality using it.
Brain without a soul.
I still like the phrase.