How AI workers earn your trust — the safety model behind every action.

An AI workforce is a team of AI workers — each one a specialist with its own job, own opinion, and own approval rules. Crewmerce ships AI workers covering commerce, creative, logistics, integration, search, AI search, ads, analytics and strategy. Unlike a single general-purpose chatbot, an AI workforce treats each domain as its own worker with the trust ladder and a clear Receipt on every action.

A chatbot answers questions and produces text. An AI agent — a worker, in Crewmerce vocabulary — additionally executes real actions against external systems under your approval rules, with every action backed by a clear Receipt. Crewmerce workers can update orders, send replies, generate marketplace-ready product photos, build campaigns, and synthesise across platforms. They prepare the work before they act and leave proof after. A chatbot cannot do that.

Every worker starts in read-only mode or suggested-action mode — it reads, prepares the work, and waits for your tap. After enough approved actions of the same kind, the worker becomes eligible for light approved automation on that capability. You grant it explicitly, per capability — never the whole worker at once. Broader automation is a reserved ceiling and never applies to refunds, bulk pricing changes, or anything financial. The full four-step trust ladder (Read, Prepare, Approve, Automate) is defined in the vocabulary.

A Receipt is the clear record left after every worker action. It carries what was found, what changed, why it changed, when it happened, and which worker did it — plus the rule that triggered the action and a plain-English rationale. Receipts are immutable, queryable from the dashboard, and exportable for audit. The action record is the searchable log of every Receipt your workforce has ever produced — see the vocabulary for the full definition.

Every AI-generated image and rewrite is labelled and trackable, with upcoming EU AI transparency expectations in mind. Images carry two independent labelling signals — a machine-readable provenance manifest plus an invisible pixel-domain watermark — so a downstream tool can verify the AI origin even if one signal is stripped. Text rewrites carry a first-exposure disclosure surfaced before publish.

The EU AI Act's transparency provisions take effect mid-2026 and aim to require: (1) machine-readable labelling on AI-generated content; (2) clear human-facing disclosure when content is AI-generated; (3) transparent logging of decisions affecting users; (4) opt-out paths from training-data use. Crewmerce is built with these expectations in mind: labelling default-on, disclosure inline, action record as the transparent log, and zero use of customer data for training — past, present, or future.

Yes. Crewmerce ships an AI client integration server, so any compatible AI assistant can hire and operate the workforce through a chat — the user-facing entry point is one universal call into the workforce. Every Receipt, every worker decision, every trust-step promotion still flows through Crewmerce. The integration is a transport; the worker is the same worker, same chat, same Receipt format.

Yes — Crewmerce uses a one-call integration pattern. The dashboard chat, the AI client integration server, Slack, the browser extension, and every future surface all wrap a single call into the workforce. There is no per-worker-per-connector matrix. Add a new connector? It calls the same workforce entry point. Add a new worker? Every connector picks it up automatically.

Three layers protect you. (1) Before any action: workers in read-only mode and suggested-action mode never act without your approval — proposals carry the proposed action, the rule and the cost preview, so you catch mistakes before they ship. (2) After any action: every Receipt is auditable; the action record lets you find exactly which worker did what, why, and at what cost. (3) Reversal: most worker actions are reversible (order edits, product updates, replies) — the worker proposes the undo on request. Financial and bulk actions stay approval-first indefinitely.

Crewmerce uses three work levels — Read (free), Draft & prepare (included in your subscription), Generate & automate (heavier work, shown as a cost preview before it runs). Every worker turn that crosses into heavier work shows you the expected cost before you approve. Your monthly cap stops total spend from being a surprise. You cannot accidentally bulk-generate yourself into a cost spike.

Today, no — the worker registry is curated. The roadmap (post production launch) adds workspace-private custom workers built from a worker template: system prompt + capability pack + default trust-step + approval rules. The same lifecycle that ships pre-built integrations also governs custom worker promotion: a workspace-private worker that proves competence across multiple shops can graduate to a public pre-built listing.

Crewmerce runs on a managed AI execution layer — leading AI models, selected per task. We manage the providers so you do not have to. The right AI capability is picked for each job. The Receipt records what was decided, so you can see what happened without thinking about which model ran.

Other tools ship one AI agent for one job — a no-code agent builder, a customer-support agent, a conversational shopping AI. Crewmerce ships an AI Agents Crew — specialised workers across the whole shop, not one agent for one job. Three differentiators on top: (1) a clear Receipt on every action — black-box AI does not give you that; (2) the four-step trust ladder with explicit owner approval at each step — competitors default to autonomous; (3) one universal chat across every surface — competitors ship per-surface integrations. The metaphor moat is the Receipt; the model moat is the workforce.