DocsBot's FAQ has a rule that doesn't look much like a selling point: when the monthly AI credits run out, the customer-service bot stops answering and shows visitors a friendly error message. The product turns a company's existing documents, websites, and help centers into an AI agent that can answer questions from customers and internal teams; once the credits are gone, it shuts up first. Business owners can upgrade their plan, turn on auto-add so credits are added automatically, or wait for the next calendar-month reset. [1]
Putting this rule at the start is worth it because it exposes the most awkward part of this business: you're selling a support agent that is always online and never gets impatient, but it will stop answering when credits run out; at the moment it stops answering, the brand standing in front of the visitor is still yours.
It Takes Over What You've Already Written
What DocsBot does is itself not complicated: it turns material a company already has into an agent that can answer customer and internal-team questions. The content sources listed on the official site range from websites, documents, and cloud drives to Notion, Confluence, and Zendesk help centers; after import, the system processes and indexes them automatically. [1]
The processing step is cleaning and splitting: first it finds the most relevant passages in the imported material, then the model answers based on them. Then comes the part people have to do: choose which sources to use, test which questions get answered off base, decide how to write custom instructions, and whether to let it take action.
Taking action is another layer. AI Actions lets the agent do more than explain the next step: it can check account status, create tickets, book demos, look up invoices, and similar tasks. Permissions can be very granular: many MCP servers expose read and write tools, and the official recommendation is to enable only the minimum necessary ones; actions can be limited to the user in the current conversation, and conversation history and tool activity are kept on file for auditing. [2]
Image: source DocsBot actual image source[3], accessed 2026-11-10, original interface retained.
credits Spread 'How Expensive the Model Is' onto Every Sentence
An ordinary chat message starts at 1 AI credit on the 1x model; switching to a larger model, or having the agent execute actions, run the skill builder/tester, or handle longer and more complex requests, can cost more—how much exactly is deducted depends on the model and task complexity. [1]
Need more credits? Add them: every 5,000 monthly credits costs $49/month, an extra bot is $19/month, and an extra 10,000 source pages is $29/month. auto-add adds 5,000 credits each time, and the account keeps the new monthly cap until you lower it; if you lower it before the billing cycle ends, the unused balance becomes account credit for future invoices. [4]
Free is $0; Personal is $49 a month for one bot, 5,000 source pages, and 5,000 monthly credits. Standard at $149 a month is the dividing line: Skills, ticket escalation, Stripe customer-service actions, and the Help Scout integration all start here, and the support bot can keep moving after it answers the question. Business is $499 a month, lifting bots, source pages, credits, and seats another notch, and adding client-side PII redaction and an unbranded chat widget so visitors don't see the DocsBot logo. Pricing[5] The top of the page also says 14-day money-back guarantee.
One comparable product in the same category is SiteGPT: Starter is $59 a month, or $468 a year, and the self-serve plan starts with a 7-day free trial and requires a credit card. SiteGPT Pricing[6] Both are selling 'turn your knowledge base into customer service,' but with different billing units: SiteGPT splits by number of messages per month, with the message allowance converted differently depending on the model; DocsBot splits by credits, with model size, actions, and request complexity all counted toward consumption. Which of the two is cheaper depends on how long and convoluted the questions are; without hands-on consumption testing, this article draws no conclusion.
This Pricing Took a Year to Grind Out
In a retrospective signed by Aaron, he writes that DocsBot crossed $1M ARR in February, about four months later than planned; growth was flat for most of Q3 and Q4 2025. The pricing overhaul took nearly a year of planning, testing, and communicating with customers, and only after it officially launched did it push the business past that line. [7]
This one-year duration is itself a confession. For a self-serve subscription company, changing prices isn't just changing a few numbers; it changes who comes in, who stays, and who pays more. His moves follow that direction: remove the consumer-tier plan, put more value into Standard and Business, tighten the sales process for qualified accounts; onboarding was also reworked so new users get to their first useful answer faster. The agency/implementer partner line is still early, and a feature as seemingly small as per-bot RBAC is an entry requirement for a partner managing multiple clients and teams.
In another post, he also put the decline in signups on the table. He writes that DocsBot's signups actually fell this year; rising ARPU and customer lifetime value only 'barely' made up the gap in signup volume. He says explicitly that he doesn't know how long improvements in customer quality can cover for weakening acquisition. [8]
He also disclosed the channel layer: comparing the first trackable source for users who completed paid checkout in months 1—8 each year, Search fell from 48.5% in 2024 to 40.6% in 2026, still tied with Direct as the largest; Direct rose from 35.4% to 40.6%, but he cautions that Direct mostly means referral signals weren't fully tracked. What surprised him was the self-reported source of people who booked demos: word of mouth and referrals accounted for 20.5%, tied with Google/Search. He sees it as evidence that something earlier—the product, trust, or a specific result—had been done right. [8]
The $1M ARR figure is the founder's own account; there are no audited financials, and no complete cost, gross profit, or model inference cost. What can be confirmed is how it collects money and how it cuts off the supply; what can't be confirmed is profit. A company that writes 'no credits, no talking' into its documentation is effectively handing the bill to customers to calculate themselves; and its own books—signups falling, average revenue per user rising—say more about where the next step gets hard than the $1M ARR number does.
[1] DocsBot - AI Agents for Business | AI Customer Support & Team Automation
[2] AI agents that take action | DocsBot
[3] DocsBot actual image source
[4] Add-Ons FAQ - DocsBot AI - Add-Ons FAQ
[5] Pricing & Savings - DocsBot
[6] SiteGPT - Make AI your expert customer support agent
[7] We finally hit $1M ARR in February
[8] We Had Fewer Signups. Better Customers Barely Made Up for It.