An agent that answers from your content, not the internet.

Most “AI support” is a chat window bolted onto a general model. It is fluent, and it is guessing. A visitor asking for *your* refund window does not want a plausible paragraph. They want the policy you actually run.

AssistLoop’s agent is a support front line trained on your own content. That is the whole product, stated plainly: it answers from what you gave it, around the clock, and it stops when it should. Live chat and tickets are what happen after that.

Four sources cover where support knowledge usually lives: uploaded PDF, DOCX or TXT files (up to 42MB each), a crawl of your website from one root URL, pasted text, and pinned question-and-answer pairs for the answers that cannot be paraphrased.

It does not invent a missing policy to sound helpful. If the answer is not in the sources, it says so, or a person takes the thread. It also does not attach source links to its replies — the customer gets a plain-language answer, not a bibliography. How you feed it, prune a crawl, and retrain a source that changed is on the training page. Which model reasons over that material, and what each reply costs in credits, is on model selection.

The objection to every support bot is the same: what happens on the question it cannot help with. Here there are two exits, and they are not the same job.

A person takes over the same thread when the customer is still waiting — refunds that are actually complicated, someone who is already angry, anything that should not be left with a model. That is human handoff, on every paid plan. You can also turn the AI off entirely and run the widget as a live inbox.

A ticket is filed when the work will outlast the chat: a bug, an investigation, a refund that will take two days. Native tickets sit next to the inbox, with an owner and a private link for the customer. They are not a HubSpot sync.

Install is a single script tag, or an official plugin on WordPress, Framer, Bubble or WHMCS. The widget can look like yours: name, logo, colours, greeting. On the free plan you can change every appearance setting and watch the preview; only the agent name and interface language persist until you upgrade.

Answering from documents is the default. If a question needs a live system — order status, a Calendly slot, a lead form — that is an agent action. You wire the endpoint; the agent gathers what it needs in the conversation and calls it. Actions are how the front line stops being a search box. They are not required to go live.

Free is $0, 150 message credits a month, no card. You can train an agent and put the widget on a site. Paid plans start at Basic $29 (3,000 credits), then Pro $130 and Business $390. Annual billing trims the monthly rate. See pricing for what each tier includes.

Each AI reply spends credits. The cost per message is shown next to the model before you pick it. Credits reset each month and do not roll over. Handoff and native tickets are paid-plan features. Extra credits, agents and seats are add-ons.