Frequently asked questions
Answers to the most common questions about Bookbag — what it is, how it stays accurate, what it costs, the channels and languages it supports, security and data handling, bring-your-own keys, and migrating from Chatbase.
View as MarkdownQuick answers to the questions we hear most. If you don't find what you need here, the section docs go deeper — start with Getting started.
About Bookbag
Bookbag is an AI customer-support platform built for ecommerce. You connect your store's knowledge — help center, product data, policies, past tickets — and Bookbag builds an AI agent that resolves customer questions 24/7 across chat, email, and social. Every answer is grounded in sources you control. See the introduction.
A generic chatbot guesses from its training data. A Bookbag agent answers only from your real data using retrieval-augmented generation, with citations back to the sources used — so it resolves order, returns, shipping, and product questions accurately instead of making things up.
It can take real actions — collect leads, escalate to a human, look up an order, or call your own API. It can run a full help desk (tickets, assignment, takeover, saved views), reach out proactively with outbound campaigns, and render interactive widgets like order cards and product carousels inside the chat.
No. You can describe your agent in plain language and the builder drafts the prompt and settings, crawl your website to train it, and paste a one-line snippet to deploy. Developers get a full REST API and webhooks when they want them.
Accuracy & hallucinations
When a customer asks something, Bookbag first retrieves the most relevant chunks from your indexed sources, then asks the model to answer using only those chunks. The facts come from your data, not the model's imagination. See Response quality.
A well-configured agent says it doesn't know and offers to connect the customer with a human, rather than guessing. You set the fallback message and can pair it with the Escalate to a human action so the conversation lands in your help desk.
Yes — add a Q&A pair. Q&A sources take priority over crawled text, so for refund windows, warranties, and shipping promises, the exact wording you approve is what customers get. See Best practices.
Two ways: Analytics tracks resolution rate and escalation trends over time, and the free Support Audit scores transcripts for hallucination rate, policy violations, quality, and resolution.
Data sources & training
Crawl your website, upload files (PDFs, docs, spreadsheets), paste text snippets, or write Q&A pairs. Bookbag chunks and embeds each source into a vector index. See Data sources.
Re-crawl or re-upload when your content changes; on higher plans you can schedule automatic retraining. Stale data is the most common cause of wrong answers, so refresh sources when policies change.
It depends on your plan — from 10 sources per agent on Free up to thousands on Pro, and unlimited on Enterprise. See Plans & billing.
Pricing & credits
There are five plans — Free, Hobby, Standard, Pro, and Enterprise — each with a monthly credit allotment and its own limits on agents, seats, and channels. See Plans & billing and the pricing page for current prices.
Yes. The Free plan includes 100 monthly credits, one agent, and the website widget plus API, so you can build and test a grounded agent end to end at no cost.
A credit is the unit of usage. Each AI reply costs credits based on the model it uses, so cheaper, faster models stretch your allotment further. Your plan refreshes a monthly allotment, and you can see usage and remaining balance in billing. See Credits & usage.
You can upgrade your plan or enable auto-recharge, which tops up your balance automatically when it drops below a threshold you set. See Plans & billing.
Channels
A single agent can answer across a website widget, email, WhatsApp, Slack, Instagram, and Facebook Messenger — consistently, from the same knowledge. See Deploy & channels.
Copy the one-line embed snippet from Deploy → Website widget and paste it before the closing body tag in your theme (on Shopify, that's theme.liquid). See Website widget.
No. Free includes the website widget and API; Hobby adds Slack and email; Standard adds WhatsApp, Messenger, and Instagram; Pro and Enterprise include all channels. See Plans & billing.
Security & data handling
Only members of your workspace, scoped by their role — Owner, Admin, Editor, or Viewer. Roles control exactly what each teammate can see and change. See Members & roles.
Yes. Grounded answers come only from the sources you add, with citations you can inspect. You control the system prompt and guardrails that tell the agent what it must never do — invent policy, prices, or order details.
You set guardrails to prevent it. The Support Audit also flags policy issues like PII exposure in existing transcripts so you can close those gaps before deploying.
Yes — Bookbag has an open-source build you can run yourself, with unlimited local entitlements and no metered billing. The hosted cloud handles credits, billing, and scaling for you.
Models & bring-your-own keys
Bookbag supports multiple models (including OpenAI and Anthropic families). You pick the model per agent from a dropdown, balancing speed, cost, and reasoning ability. See Models & model choice.
Yes. With bring-your-own keys enabled for your workspace, your agents run on your own provider key and those replies aren't metered against your Bookbag credits. See Bring your own API keys.
Each model has a configured credit cost per reply. Cheaper models consume fewer credits, so routing routine questions to a fast model and reserving a larger model for complex cases keeps usage efficient. See Credits & usage.
Languages
Yes. The underlying models are multilingual, so your agent can understand and reply in many languages. Provide knowledge in the languages your customers use for the most accurate, grounded answers.
Yes — the help desk includes translation so your team can read and reply across languages. See Help Desk → Translation.
Getting started
Follow the first-agent walkthrough: create an agent, add data sources, test it in the playground, and deploy the widget — about ten minutes end to end.
Yes. The live trial spins up a working agent on your own website in under a minute — paste your URL, we crawl it, and you can chat with a grounded agent immediately, no account required.
Invite teammates by email and assign a role. Existing Bookbag users join immediately; new emails get a pending invitation that expires after 24 hours. See Invitations.
Migrating from Chatbase
Bookbag covers the same core surface — grounded agents, data sources, a playground, deployable widget, actions, and contacts — and adds ecommerce-focused features like a full help desk, outbound campaigns, interactive widgets, and a free support audit. You can rebuild your agent in minutes by re-crawling the same sources.
Yes — that's exactly what the free Support Audit is for. Paste your current bot's transcripts, get a scored report on hallucinations, policy violations, quality, and resolution, and use it as a baseline to measure the improvement after switching.
Chatbase's "Member" maps to Bookbag's Editor. Bookbag adds a distinct read-only Viewer role for stakeholders. See Members & roles.
Yes. Enable bring-your-own keys so your agents run on your provider key, just as you may already do elsewhere.