Pulson FAQ

Short answers to the questions people ask most about Pulson. For the AOP operator CLI, see the AOP CLI FAQ.

What is Pulson?

PulseAd’s AI-powered Amazon advertising platform. You talk to Pulson in natural language; it answers from scoped account data and can act on the account (budgets, bids, campaigns, keywords, and targets) when the relevant approval or automation policy allows it. See What PulseAd is.

Is Pulson the same as PulseAd?

PulseAd is the platform and product; Pulson is the AI assistant and the customer-facing experience. You may also see Growth Twin in product surfaces and older material; it refers to the same brand-facing assistant experience, not a separate product. (PULSEAD is also the main Snowflake schema — a schema, not the UI.)

How is the product organized?

PulseAd has three connected layers:

1. In-app agents — the agents users open in the Pulson app, grouped into Planning, Reporting, and Campaign.
2. The chat engine — the ai-api Action Chat graph that routes a request to a specialist and composes the answer.
3. Autonomous back-office agents — scheduled and orchestrated jobs in pulsead-agents that monitor accounts, prepare proposals, run gated automation, and feed results back into the product.

See the Agents overview for the full map.

How many agents are there?

Count agents by product surface rather than adding them into one headline number:

• 11 visible in-app agents: 1 Planning agent, 6 Reporting agents, and 4 Campaign agents. The source of truth is the frontend agent registry.
• 18 scheduled/orchestrated back-office agents in pulsead-agents, covering campaign monitoring, optimization, ad scheduling, AMC, proactive keyword automation, and support reminders.

The 11 includes Overview Dashboard and Event Monitoring in Reporting, and Keyword Optimizer in Campaign. The former Planning mock entries are gone; the current Planning tab contains only News & Market Intelligence.

The back-office count reflects the current source state: the legacy four-agent weekly-report LLM pipeline was removed, and a support-reminder job was added. The in-app Pulson Weekly Report still works; it reads published content through the ai-api weekly-reports API. The ai-api chat, report, and diagnostic roles, plus the separate AOP operator CLI, are different surfaces and are not additional in-app agents.

Can Pulson change my Amazon account?

Yes, but never silently. It can propose and, where explicitly allowed, execute changes to campaigns, bids, budgets, keywords, and targets:

• Campaign Optimizer and Keyword Optimizer use a proposal-then-chat- approval flow.
• Budget Automation is the exception: it may run a multi-stage optimization cycle under an explicit auto-approval policy.
• Every Amazon Ads write is subject to role checks, approval state, brand mode, and budget guardrails.

See Capabilities for the complete read/write matrix.

Who can approve a change?

A permitted user. Viewer-role members can’t approve campaign changes (they get a 403); budget changes also pass size guardrails (≤50% / ≤$10k / ≥$1). Approval cards expire after a short window, and a request is rejected if it is stale, the account is disabled, or the brand mode does not allow the operation.

Does Pulson ever act on its own?

Yes, but only within defined gates. Budget Automation is the only in-app agent that can execute Amazon budget or bid changes without a separate chat approval for every directive. With auto-approval disabled, the cycle waits at its human approval gates; with it enabled, only the configured policy can authorize the next stage.

Campaign Optimizer and Keyword Optimizer remain chat-approval-gated. AMC Builder, Planning, and Reporting are read-only with respect to Amazon Ads. Autonomous budget writes currently reach one whitelisted pilot brand; for other brands, the system calculates and records the decision without sending the mutation to Amazon. The support-reminder job is internal support automation, not an Amazon Ads write path. See Campaign agents.

What happens if something fails?

The system is designed to fail closed. An error or timeout should produce a visible error, fallback, or pending/review state rather than a fabricated success. Approved mutations are recorded in execution history with their status so an operator can retry, investigate, or leave them pending.

What data does Pulson see?

Amazon Ads performance through six domain analysts over Snowflake (refreshed nightly), plus off-Amazon channels that are uploaded manually. It can also read selected external market and product information, and AMC queries and results through AMC Builder. Operator context — workspace, team, country, permissions, connected profiles, and approved preferences — is scoped to the request. See the data model and data & coverage.

Which AI models does Pulson use?

The exact model depends on the surface:

• ai-api chat, report, and planning flows: the OpenAI GPT-5 family.
• Search-term segmentation: Gemini 2.5 Flash-Lite, a specialized exception in ai-api.
• Autonomous pulsead-agents jobs: AWS Bedrock Claude models — Haiku 4.5 by default, Sonnet 4.6 for writer and decision agents, and Opus 4.8 for weekly- report summary generation.

Models run behind permission checks, structured outputs, deterministic rendering, and approval gates; a model does not receive permission to write to Amazon merely because it generated a recommendation. The full map is on the AI models page.

Why did it ask me to pick a team or profile?

Almost everything runs within a workspace/team + country scope, against one active profile. Without a resolved scope, data calls fail closed (empty or not-permitted). See the reference.

Is my data isolated from other brands?

Yes — Pulson is multi-tenant and fails closed when scope is missing. It never shows one customer’s real names to another, and these docs use placeholders only (Brand A, ASIN1234).

What’s “PSM”? Is it still used?

PSM (psm_chatbot.py, /pulsy/chats) is the original, now-frozen chat engine. The current live chat path is Action Chat: action_chatbot.py and the /pulsy/action-chats route. Its graph classifies a request as a query, external insight, Amazon Ads action, or internal task, then routes it to the appropriate specialist before composing a response. New chat capabilities should be added to Action Chat — see the name map and what is deprecated.

Are reports generated automatically?

There are two different reporting paths:

• On-demand generators create daily briefs, weekly reports, and monthly reports through the Dashboard Report Agent. They fetch scoped data, plan structured report directives, and render predictable KPI cards, insights, charts, and tables.
• Pulson Weekly Report is a managed, operator-published report. The customer-facing endpoint serves already-published content; it is not the old autonomous weekly-report pipeline.

Reports and dashboards are read-only with respect to Amazon Ads. An action card in a report is a recommendation, not an applied change.

Can I ask Pulson for strategy without changing anything?

Yes. Ask for a diagnosis, explanation, market or competitor context, a report, or a simulation without asking to apply a change. Pulson can return read-only analysis and recommendations. A request to actually modify an Amazon Ads entity is routed through the action path and its approval or policy gates.

Does Pulson manage non-Amazon ad platforms?

The product is designed around Amazon Ads data and controls. It can use external market and product sources for context, but that does not mean it can manage arbitrary ad platforms.

Where can I learn the full product map?

Start with the overview, then see agents, capabilities, AI and models, and what is deprecated.