Agentplace

Agentplace publishes each agent at its own URL, where the model picks the content and components a visitor sees. The vendor describes responses that arrive as plans, tables, charts, product cards, forms, maps and documents rather than chat text, and a changed requirement revises the result and the page together: its catering example updates dishes, portions and the estimate when a customer adds guests.

There are two deployment shapes on the homepage: chat and voice added to an existing site, where a request opens the agent across the full page, and a standalone agent that is itself the destination. Phone, WhatsApp and Telegram are listed as separate channels that need the relevant provider configured first, and the vendor is explicit that web voice does not automatically provide a phone number.

Free is a standing $0 plan with 2,000 AI credits, and Pro starts at $29 a month with 15k monthly credits plus one custom domain. The Pro card offers larger credit steps up to 235k; those steps carry no printed prices beside the entry figure. There is no monthly/annual control on the page, and Pro is labelled Billed monthly.

The docs then spend a whole page warning against arithmetic on that allowance. A credit is not a message, one request that includes research, tool calls and revisions is not a short answer, and a visit can start billable model work before anyone types. Connected services and scheduled work can charge separately. The counterweight is the free model: GPT-6 Luna is free in the Builder and the agent, and an agent at zero credits switches to it automatically rather than going dark.

Business apps are reached through Composio's catalog of 1,000+ integrations, with Google Calendar, Gmail, HubSpot, Salesforce, Stripe, Shopify, Zendesk, QuickBooks and vertical tools such as Clio Manage and Workiz named as examples. The same page refuses to treat the catalog as a guarantee: availability depends on supported authentication, the provider's API, your account plan and the enabled tools, and you are told to have the Builder check the specific action.

MCP is the more interesting surface. A published agent exposes an MCP connection for compatible clients, so Claude or ChatGPT can call the business's agent, subject to its authentication and access mode. In the other direction the agent can delegate part of a task to a configured tool or another agent. A public REST API for managing agents is not documented anywhere on the site.

Publishing offers Public, Restricted and Private modes, with Restricted matching an allowed email address or email domain and an optional Require sign-in switch on Public agents. The same page draws a line most vendors leave blurred: sign-in identifies a visitor and does nothing else, so payments, audience limits and app authorization are separate configuration.

Security disclosure stops at prose. Data is encrypted at rest and in transit, infrastructure spans AWS, Google Cloud, Azure and AI providers depending on the model used, and no SOC 2 report, ISO certificate or DPA is published; buyers are told to email for security documentation. The vendor states it does not train models on customer data, then says written guarantees of that are available only through an Enterprise contract, because its own providers' practices differ. A complete export of an agent's stored data is Enterprise too, while deletion on request is said to take effect immediately.

The operator is Agentplace, Inc., with Uladzislau Yanchanka and Eugene Sorokin named as founders on the company reference page and an address at 131 Continental Dr, Newark, DE on the privacy policy. Stripe handles subscriptions, Amazon Cognito handles sign-in, and Sentry, Amplitude and Google Analytics appear in the provider list.

The product's public history runs through three Product Hunt launches from November 2024 to March 2026, plus VentureBeat coverage. Press logos for VentureBeat, USA Today and Fast Company sit at the top of the homepage. One date mismatch is worth noting: the privacy policy is effective 13 October 2026 while the Terms of Service still carry an effective date of 01-06-2023 and read as website terms, granting a licence for "personal and non-commercial purposes" and saying nothing about subscriptions or cancellation.

The fit the vendor argues for is a customer task rather than a lookup: incomplete or changing requirements, business rules that shape the recommendation, and a result the customer wants to inspect and revise. Travel advisors, caterers, contractors, accounting practices, retailers and consultants selling analysis as a digital service are the named buyers, and owners on a paid plan may build agents for paying clients and set their own fees.

Its own limitations page rules out two groups. A simple FAQ or a fixed trigger-and-action workflow is a better fit when nobody needs a personalized experience, and anyone needing to run the stack themselves is out: the platform is not open source or self-hosted, and syncing generated agent code to GitHub does not change that. Teams that must see an audit report before signing will be in an email thread, not a self-serve checkout.