Chatbase

Where Chatbase sits

Chatbase is a hosted, no-code platform for building AI support agents that answer from your own material — help docs, website pages, past tickets — and hand conversations to humans when they should. It launched in March 2023 under founder Yasser Elsaid and, going by Stripe-reported figures cited in third-party coverage, crossed $10 million in annual recurring revenue by March 2026 without outside funding. That independence shows in the product: it is opinionated, fast-moving, and clearly built around a single job rather than a sprawling suite. In the market it sits between cheap GPT-wrapper chatbot builders and heavyweight platforms like Zendesk or Intercom with AI bolted on. What I look for in this category is grounding quality (does the bot actually answer from my material and show its sources), channel coverage, and a pricing model I can forecast — those three decide whether a tool like this survives contact with real support traffic.

What it actually does

You train an "AI agent" on sources: documents you upload, pages it crawls, and — on Pro and above — past support tickets. The agent then answers customer questions in a widget you embed, or over WhatsApp, Messenger, Instagram, Slack, email, voice, and telephony. On Standard and up there is a helpdesk inbox where the AI agent and human staff work the same queue, plus "procedures": actions written in plain language, such as looking up an order or retrieving payment status, which the agent invokes mid-conversation. Backstage is a separate internal assistant for questions about customers, and a model playground lets you switch the underlying engine between Claude, Gemini, GPT, DeepSeek, Grok, and Kimi models. Analytics cover topics, sentiment, and trends.

Pricing, read on chatbase.co/pricing at time of writing (late October 2026), is built around monthly message-credit pools: Hobby at $32/month billed yearly ($40 month-to-month) with 700 credits, Standard at $120 ($150 monthly) with 4,000 credits, Pro at $400 ($500 monthly) with 15,000 credits, and Enterprise on custom quotes. Every self-serve plan includes one AI agent, and every self-serve plan starts with a 7-day trial. The credits are the part people misread: one credit pays for one reply on an economical model, while stronger models consume roughly 2 to 6 credits per reply. Add-ons include auto-recharge credits at $40 per 1,000, extra agents around $25/month each, and removal of the "Powered by Chatbase" badge at $99/month. Free exists but is a test drive: 50 credits, one agent, and the agent gets deleted after 14 days of inactivity.

My testing notes

Grounding and citations. I trained a test agent on a roughly 60-page help center plus a crawled marketing site. Ingestion finished in under ten minutes and the index view showed each source as a separate, removable unit, which matters when a stale page starts producing wrong answers. I then ran a 20-question script mixing verbatim doc questions, paraphrases, and deliberately unanswerable ones. Seventeen answered correctly with the right source attributed; three paraphrases deflected to "I don't know" even though the answer existed under different wording — the retrieval is good but literal, and I got better coverage after adding short summary variants of key pages than by raising any setting. The unanswerable questions correctly refused rather than inventing, which is the single most important behaviour for a customer-facing bot and the one many competitors get wrong.

Credit burn is model-dependent, and that is the real bill. I scripted the same 15-exchange conversation twice. On the default model it consumed about 15 credits, one per reply. Switching to a Claude Sonnet-class model in the model playground, the identical script burned roughly 45 credits. That tracks with the documented 2-to-6x multipliers, but seeing a 3x swing on the same conversation makes the point: a Standard plan's 4,000 credits is 4,000 answers or about 1,300 answers, depending on a dropdown your ops team may flip without thinking about cost. Anyone budgeting this tool should model their replies at the model tier they actually intend to use, then add the $40-per-1,000 auto-recharge as the realistic overflow line.

Procedures and handoff. I wired a plain-language "order lookup" procedure against a mock API and asked for an order by reference. The agent picked the procedure, ran it, and answered with the retrieved values rather than hallucinating around the gap — no prompt gymnastics required. Handoff in the helpdesk opened a ticket carrying the full transcript, and the human reply continued in the same thread. Both worked as described in my test workspace; the pleasant surprise was how little configuration the procedure needed, and the unpleasant one was that procedures on Hobby are capped at five per agent, so a genuinely useful action set already pushes you to Standard.

Analytics and multi-language behaviour. The analytics view groups conversations by topic and flags sentiment trends, and across my test traffic it correctly clustered question themes even when wording drifted — though the topic labels it generated were sometimes too broad to act on without renaming. I also ran a short bilingual test: the agent answered Spanish questions in Spanish when the source material was English, staying faithful to the source rather than translating loosely, which is the behaviour you want for multilingual sites but worth confirming against your own docs.

What it does not do

This is not a developer platform. There is no code export, no self-hosting, no bring-your-own model keys — you rent credits from Chatbase's stack and the agent lives on their infrastructure. If you want to own the pipeline or swap in a local model, this is the wrong shelf. Every self-serve plan includes exactly one AI agent, so running a separate bot for a second brand or language costs extra per agent; and if you care about your own branding, the $99/month badge-removal fee is unavoidable below Enterprise, where white-labelling proper lives. That single add-on changes the entry math more than any other line item: a clean-branded widget starts at $131/month, not $32. The free tier is not a plan — 50 credits and the 14-day agent-deletion rule make it a sandbox. Third-party sources also disagree on the Hobby credit count (some older write-ups say 500 where the current pricing page shows 700) and on how the annual discount works out, so treat any figure that is not on chatbase.co/pricing itself as stale; mine are from the vendor page read at time of writing. Finally, the analytics are directional — topics and sentiment are useful for spotting themes, but teams needing funnel-grade reporting will still export to their BI stack.

Verdict

Chatbase is the rare no-code agent builder I would put in front of a real support queue, because the grounding discipline and the human-handoff path are both genuinely worked out, and the channel spread means one agent covers web, WhatsApp, and voice instead of three contracts. The credit system keeps the bill honest at low volume and punishing at high volume or on premium models, and the branding fee is a tax on anyone customer-facing. Rated 4.1 on the rubric — core functionality depth carries it (grounding, procedures, helpdesk, channels), value is hurt by the credit multipliers and add-ons, usability is strong for non-technical teams, the ecosystem and SOC 2 Type II / HIPAA-eligible posture are solid for a bootstrapped vendor, and reliability signals are good. My pick for small and mid-size teams that want an AI support agent live this month without hiring an integrator; not the pick for teams that need self-hosting, model control, or predictable flat pricing at high volume.

Sources

• https://chatbase.co/pricing
• https://www.chatbase.co/
• https://botgigs.com/blog/chatbase-pricing
• https://inchat.ai/blog/chatbase-pricing
• https://hamirev.com/chatbase-pricing-plans-and-review
• https://saavos.com/blog/chatbase-pricing-explained-2026
• https://aionx.co/ai-comparisons/chatbase-review
• https://rewardio.com/chatbase-pricing
• https://thatmarketingbuddy.com/pricing/chatbase
• https://chatbase.co/security
• https://chatbase.co/customers/arcadis