Meet Loam's four AI collaborators.
Carl architects. Diana designs. Anthony builds. Abish verifies. Each brings a distinct working style to the same project record—with no hidden human operators.
They are AI collaborators with synthesized voices and images. Project records are account-scoped and user-controlled. Security documents the storage, deletion, and infrastructure boundaries.
Four role-specific AI working contexts.
Carl, Diana, Anthony, and Abish run on Anthropic's Claude. Each has a defined responsibility, a role-specific record, and a different way of challenging the work. There are no hidden human operators behind the profiles.
The team structure separates architecture, interface quality, implementation, and verification. The shared engineering record keeps those roles aligned on the current decision, finding, and open work.
Meg answers the phone. She is not one of the four.
Meg is an AI intake context with a synthesized voice and image. Her shipped pickup is “Loam, front desk — this is Meg.” She verifies callers before disclosing account information, records the request in the caller’s words, and routes supported work.
She does not perform engineering tasks. If a request falls outside the current phone workflow, she records the boundary instead of inventing a capability.
Caller recognition does not bypass verification. Account, order, and work details remain behind the identity gate.
Coherent working identities replace brittle rule piles.
A long list of local instructions can compete with itself. A coherent role gives the model one global working context: an architect who demands evidence, a designer who protects hierarchy, a developer who owns implementation, or a reviewer who checks claims against source.
Each role has a stable responsibility, review posture, and working record. That gives the model a consistent decision frame across sessions. Loam calls this internal design pattern behavioral compression; it is a product hypothesis, not a published benchmark.
"Stable context is an engineering choice."
The record updates the working context.
The “What he remembers” and “What she remembers” rows in the cards above describe the kind of entry each role keeps, not current status. Learning-active sessions can add role-specific decisions, risks, and observations to the account-scoped record; those entries stay in the record rather than on this page.
Relevant entries can be recalled into later work, and the system can surface a cross-team signal when multiple roles record the same issue. The source entries remain the evidence behind that summary.
This is a historical example of the system grouping related observations. It does not prove a universal advantage over another assistant; it shows how Loam organizes its own working record.
Open the same working room by phone.
Meet the Team opens a synthesized-voice session with Carl, Diana, Anthony, and Abish. The relevant account-scoped record is loaded before the call so the group can work through the decision in front of you.
Calls can load relevant project history and prior decisions from your account-scoped record. Personal context appears only when you choose to share it. You can review or delete that record at any time.
“The difference is visible when the call retrieves the right prior decision without another briefing.”
Meet the Team starts a synthesized-voice session with all four collaborators and a prioritized working set from the relevant record. Older entries remain available through recall.
When you need one role instead of the whole group, use Call Team: Carl for architecture, Diana for design, Anthony for implementation, or Abish for QA. It opens a focused synthesized-voice session with that role.
The memory that makes these conversations possible is yours to see and delete. What Loam remembers about you, and how to control it.