Chatbase

Operate Chatbase AI chatbots from your flows. Agents send grounded chat messages, create and retrain chatbots from raw text, refresh training knowledge when documentation changes, and pull captured conversations and leads into CRM and reporting systems.

8 actions API key available

Chatbase website ↗ Platform Documentation ↗ Capability data verified 2026-07-27

The help center publishes a new release with changed refund and pricing policies

The flow assembles the refreshed source text from the updated documentation

The agent diffs the new corpus against the current training text: sections removed, policies changed, size anomalies that suggest a bad export

A support lead approves the retrain with the removed and changed sections highlighted, because this text becomes what the public bot tells customers

Update Chatbot Data replaces the training knowledge, and retraining runs asynchronously on Chatbase

Send Chat Message runs the regression set: known questions with known correct answers, confirming the bot still gets them right

The verification results post to the support channel, pass or fail, closing the loop

What This Integration Enables

A chatbot trained on your documentation has one honest weakness: it is only as current as its last retrain, and it will not tell you when it has gone stale. It answers with the same confidence the day after your pricing changed as the day before. Chatbase makes building the grounded bot easy; this connector makes operating it disciplined. FlowRunner agents run the full lifecycle around the bot: refreshing its knowledge when the source material changes, testing it like software before and after, and harvesting the two assets it produces, conversations and leads, into the systems that act on them. - Send grounded messages with full conversation history, model and temperature overrides, and conversation continuity - Create chatbots from raw text and replace their training knowledge as documentation evolves - Adjust persona, instructions, greeting, and visibility without touching the trained knowledge - Export conversation transcripts filtered by date and source for review and analytics - Pull captured leads with name, email, and phone for CRM sync The division of labor is deliberate. Reading conversations and syncing leads runs continuously at machine speed. The operations that change what the bot says to the public, retraining and settings changes, pass through a [human-in-the-loop](/concepts/human-in-the-loop) review, the same way customer-facing copy does.

Without FlowRunner

The bot answers from last quarter The docs moved on and the widget still cites the old refund policy, confidently

Leads die in the dashboard Contacts captured by the bot sit in Chatbase, unexported and unworked

Nobody reads the transcripts The clearest record of what customers actually ask goes unread

With FlowRunner

Retraining chained to doc changes Knowledge refreshes are reviewed, applied, and regression-tested as one flow

Leads reach the CRM in minutes Every captured contact lands in the pipeline while the conversation is still warm

Transcripts become roadmap input Conversations are mined weekly for the questions the bot could not answer

Use Case Scenarios

The retrain that ships like a release

When the docs change, the flow does not just push new text at the bot. It diffs, gets a support lead's approval on what is changing, calls Update Chatbot Data, waits out the asynchronous retrain, and then interrogates the bot with Send Chat Message across a fixed regression set: the twenty questions customers ask most, each with a known correct answer. Only a clean pass closes the change. The bot gets the same discipline as a production deploy, because to customers, that is exactly what it is.

The lead form that feeds the pipeline

Hourly, Get Leads pulls the contacts the bot captured, and each becomes a [HubSpot](/integrations/hubspot) contact with the capture date and source attached. High-intent signals, a phone number provided, an enterprise domain, post immediately to [Slack](/integrations/slack) so sales reaches out while the visitor still remembers the conversation. The bot was already collecting these names; now they go somewhere.

The transcript mine

Weekly, Get Conversations exports the week's chats into [Google Sheets](/integrations/google-sheets), and an analysis agent clusters them: what got asked, what the bot answered well, and, most valuably, where users rephrased the same question three times and gave up. Those failure clusters become the documentation backlog, ranked by frequency. The next retrain closes the gaps customers actually hit, not the ones someone guessed at.

Human-in-Loop Highlight

Update Chatbot Data replaces the chatbot's entire training knowledge in one call. There is no merge and no partial update: whatever text is supplied becomes everything the bot knows, and the bot speaks to customers unattended, around the clock. A truncated export or a wrong file does not throw an error; it produces a public-facing agent that answers from a fraction of the truth with all of its usual confidence. That is why no FlowRunner flow retrains autonomously. The agent prepares the corpus, diffs it against what is live, and a support lead approves with the changes in view. Delete Chatbot carries the harder edge of the same gate: it permanently destroys the bot along with every conversation transcript and captured lead it ever collected, so it runs only after those assets have been exported and a person has confirmed the right chatbot ID is about to die. The bot converses at machine speed; what it is allowed to know is a human decision.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via preferred channel

Human decides

Agent resumes with decision

Agent Capabilities

8 actions

Chat

Send Chat Message Sends a message to a chatbot and returns the grounded reply drawn from its trained knowledge. Accepts the full message history with user and assistant roles for context, plus optional model, temperature, and conversation ID overrides. Responses return in full; streaming is not supported through this action.

Chatbots

List Chatbots Retrieves all chatbots in the account with their IDs and names. The discovery step for every other operation.
Create Chatbot Creates a new chatbot and trains it on supplied raw text. Training runs asynchronously; the chatbot ID returns immediately for use once training completes.
Update Chatbot Settings Updates configuration without retraining: name, model, temperature, base instructions and persona, initial greeting, suggested messages, and visibility. Only supplied fields change, applied asynchronously.
Update Chatbot Data Replaces the chatbot's raw-text training knowledge and retrains it. The action that keeps answers current when documentation changes, and the reason this page's human gate exists.
Delete Chatbot Permanently deletes a chatbot with all of its training sources, conversations, and leads. Cannot be undone, and runs only after export and explicit confirmation.

Conversations

Get Conversations Retrieves a chatbot's conversation history with messages, filtered by date range and originating source (API, hosted iframe, or website widget), paginated. The raw material for analytics, review, and transcript archives.

Leads

Get Leads Retrieves the contacts captured by the chatbot's lead form, name, email, and phone, with date-range filtering and pagination. The bridge from widget conversations to the CRM.

Frequently Asked Questions

What can FlowRunner do with Chatbase?

FlowRunner agents can run Send Chat Message, List Chatbots, and Create Chatbot in Chatbase, plus 5 more actions.

Does connecting Chatbase to FlowRunner require OAuth?

No. Chatbase connects to FlowRunner with an API key, no OAuth flow required.

Can Chatbase trigger a FlowRunner workflow automatically?

Chatbase doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.

Start building with Chatbase

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us