Short answer
Chatbase is worth investigating when a small business has recurring support questions, maintained reference material and a person who owns escalations. It is a weaker fit when policies are scattered or nobody can check whether the answers are correct.
This is official-source research checked on November 21, 2026, not a hands-on deployment. We have not trained an agent, purchased a plan or measured resolution rates. Ratings are withheld, and product links are direct rather than affiliate links.
Answers, actions and escalation
The [product overview](https://www.chatbase.co/) describes agents grounded in supplied knowledge, actions in connected systems and handoffs carrying conversation context. These are vendor-described capabilities, not observed reliability. Begin with answering questions before allowing an agent to change customer records.
The important distinction is between explaining a cancellation policy and canceling an account. The second action requires identity checks, explicit authority and a recovery process. Keep those as separate acceptance criteria even if both appear in the same conversation interface.
Understand the credit budget
The [current pricing page](https://www.chatbase.co/pricing) lists Basic with 4,000 monthly message credits and Business with 15,000. It shows $150/$120 and $500/$400 monthly/monthly-equivalent figures alongside annual billing. Verify the billing selection and total commitment before checkout. The listed seven-day trial is not evidence of an ongoing free plan.
Auto-recharge is listed at $40 per 1,000 credits; branding removal and extra agents are separate charges. Credits are not a count of resolved customer issues. A long exchange, retries and the selected model can affect the useful capacity of a plan. We did not verify a live usage meter.
Proposed trial: a deliberately incomplete policy
Prepare a fictional shop policy covering delivery regions and standard returns. Leave international delivery unanswered. Write the correct responses before loading the material. Start with five answerable questions, then five questions where clarification or a person is required. These are proposed test cases, not results from Chatbase.
Include a customer who asks for an exception, an outdated promotion and a request to reveal another customer's order. The desired outcome is not always an answer. Sometimes it is a clear statement that the source does not establish eligibility and an appropriate handoff.
Inspect the conversation from the human agent's side. Can the person see what was asked, what the bot promised and why the customer needs help? A bot that merely says it has escalated has not demonstrated that anyone received the issue. Use the [support-bot escalation exercise](/articles/test-ai-support-bot-unknown-answers) to make this check repeatable.
Compare against your existing support process
Use the same question set with your current help center and saved replies. Track correct resolutions, unsupported promises, human repair time and actual usage. A quick response that produces a second ticket has not saved the team work.
Compare the operating model with [Intercom Fin](/tools/intercom-fin) if you already use a helpdesk, or [DocsBot](/tools/docsbot-ai) for a knowledge-answering shortlist. These are candidates to investigate, not tested winners. Confirm deletion, retention and access requirements before loading customer conversations.
Verdict
Shortlist Chatbase after your policies are current and a human escalation owner is assigned. Pilot one narrow question category, review every response and estimate cost from accepted resolutions. Expand permissions only after the basic answer-and-handoff workflow works reliably for your team.