Account Lead

The memory layer. Every decision has context.

Maya holds the brand's decisions, preferences, goals and constraints. Agents read it before they answer or act, and anything an agent proposes waits for an owner to confirm.

How Maya works
Leads withWhat was decided before, and when it was decided.
NeverPresent a recalled decision as a current fact, or repeat a question the record already answers.
Why this exists

Lessons paid for, then forgotten. Every cycle starts from zero.

The same lesson, paid for twice.

TikTok tested in August at $72 CAC against a $65 cap. Killed after 3 weeks. Six months later, somebody pitches TikTok again. The budget gets re-burned.

Context dies between meetings.

Monday's insight is gone by Thursday. The constraint mentioned last month never makes it into this month's plan. Each conversation starts from scratch.

Agency handoffs lose the why.

Account managers turn over. New hires take three months to find the existing decisions. Brand-finance reconciliation history walks out the door with the previous team.

Models without preferences.

Generic LLMs do not know the CAC ceiling, the channel constraints, or the gradual-change preference. Every prompt re-litigates the basics.

What Maya ships

Five capabilities. One memory layer.

Each capability maps to a specific job: hold the brand history, pick up decisions from chat, gate agent writes through a person, keep each brand apart, and serve the rest of the agents.

Brand memory across every cycle

Conversations, decisions, and outcomes preserved with the context that made them. Searchable across months, not just last week.

  • Decisions stored with the why, not just the what
  • Outcomes linked back to the original call
  • Recalled on demand, with the source of each fact attached

Example. Example. Past decisions on a channel resurface when it comes up again

Auto-extracted decisions and constraints

Decisions, preferences, goals and constraints get pulled from in-app chat and Ask, and proposed as memory candidates.

  • Up to three facts picked up per conversation
  • Decisions, preferences, goals, constraints, lessons and facts
  • Each candidate keeps the conversation it came from

Example. Example. A ceiling stated in chat is proposed as a memory to confirm

Human-approval workflow on writes

No agent writes to memory silently. What an agent or a member proposes waits as a candidate until an owner or admin confirms it.

  • Every agent write proposal is reviewable before commit
  • Owners and admins confirm, edit or dismiss in chat or on the Memory page, with undo
  • Audit trail of who approved what and when

Example. Example. A candidate from Tuesday's chat confirmed, with the person and time recorded

Per-brand isolation

Memory is scoped to each brand. Agencies running multiple brands never see context bleed across accounts.

  • Brand-scoped storage, never shared across tenants
  • Recall reads only the brand it is asked for
  • A memory can be retired or dismissed per brand

Example. Example. Two brands in one workspace, each recalling only its own

Agent-aware

Agent runs and chat read the brand's memory before acting, and Olivia's briefs do too.

  • Chat and Ask answers start from the brand's memory
  • Agent runs read it before a step is planned
  • Olivia's briefs read stored tone and constraints

Example. Example. A brief drafted with the offers the brand has said it never runs

A typical week

How memory moves through the agents.

One workweek, hour by hour. Specific events vary by account.

Mon · 9:47 AM
Surfaced

A new chat asks about TikTok. Maya's recall brings back the decision recorded when it was paused.

Tue · 11:23 AM
Updated

A promo window mentioned in chat becomes a memory candidate awaiting an owner or admin.

Tue · 2:08 PM
Approved

The owner edits and confirms it in chat or on the Memory page.

Thu · 4:55 PM
Stored

An agent run records an observation from the week's data, which goes live as a system memory.

What's underneath

Real infrastructure. Not a chat history.

Maya runs on one brand memory store ranked by relevance, recency and confidence, with person confirmation for anything an agent proposes.

Reads from

Agent runs write calibrations, observations and measurements; chat proposes candidates for a person to confirm.

Agent runs and chat write here

Reports to

Every agent on demand, with the source of each fact attached.

Answered on demand

Collaborates with

Felix, Parker, Sam, Max, Olivia, Dana, Dex, Iris, and Remy. Reads first, writes second. The connective tissue across the whole team.

Agents read the brand's memory before acting

Built on

One brand memory store, recalled by keyword overlap, recency and confidence, with a person confirming anything an agent proposes. Per-brand isolation enforced at the storage layer.

One store, person confirmation on agent writes

Ready when you are

See Maya hold the brand history, or talk to us about your stack.

The pilot connects your accounts to every agent. Maya starts recording what you confirm from the first conversation.

Looking for something specific? Felix forecasts revenue →, Sam tests budget changes → or Parker reconciles platform claims →.