Calls become memory your assistant acts on

Every system records calls. Almost none of them can answer a question two weeks later that depends on one.

The difference is whether the conversation was turned into structured, addressable context or filed as a blob with a timestamp.

Archives versus memory

An archive is searchable if you already know what you are looking for. Memory is useful when you do not: when you say “send them the price we discussed” and something has to resolve which price, which them, and which discussion.

Heyno extracts entities, commitments, preferences, and constraints from a conversation and writes them where the rest of the system can reach them.

What gets remembered

Not everything. Small talk, repetition, and things already known are dropped. What is kept is what changes future behaviour: a supplier switch, a stated preference, a deadline, a price, a person’s role.

Every remembered fact carries a provenance link back to the moment in the conversation that produced it, so it can be checked rather than trusted.

Resolving “this”

The hardest and most valuable thing memory does is resolve deixis: this, that, them, the usual. Getting it right requires knowing what was recently salient to this particular person, which is exactly what a single continuous workspace provides and a per-app assistant cannot.

Forgetting on purpose

Memory that only grows becomes noise, and stale facts are worse than no facts. Superseded values are retired when they are contradicted, and anything can be removed on request, with the removal propagating to everywhere it was used.

Scope and privacy

Memory derived from a conversation inherits that conversation’s scope. A personal thread does not leak into a team channel; a customer’s record does not become general workspace knowledge.