Create and manage Chatbase AI agents, chat with them, and run their helpdesk, sources, and WhatsApp surfaces.
Chatbase ships in the w6w first-party pack. It declares 35 actions, 2 health checks, and the host runs its code in a sandbox that never sees the credential.
io.w6w.chatbaseChatbase builds AI agents trained on your own content — a website, uploaded files, Q&A pairs — and puts them in front of customers over a chat widget, WhatsApp, Slack, and a growing set of other channels. This app covers Chatbase’s API v2, its current, structured REST surface: create and configure agents, train them, chat with one directly (as a bot-to-bot or backend integration, not just the widget), and manage the conversations, knowledge sources, helpdesk tickets, and WhatsApp templates those agents produce.
Every agent-management, chat, conversation, source, helpdesk, and WhatsApp action here talks to API v2 (www.chatbase.co/api/v2), which Chatbase itself points integrators toward from its own legacy v1 docs. The one exception is reading captured leads, which as of this writing has no v2 equivalent at all and is served from the older v1 surface (www.chatbase.co/api/v1) — both hosts resolve to the same hostname, so a single Connection and network allowlist cover both.
File-type knowledge sources (PDF/DOC/DOCX uploads) are deliberately left out: Chatbase serves that upload from a different host (files.chatbase.co) as a binary multipart body, and this sandbox’s fetch coerces every request body to a string on its way to the network — the same limitation that caps this pack’s Box and Dropbox uploads to text content. Plain-text, Q&A, and link sources (the create-source / update-source actions here) are unaffected and cover the rest of Chatbase’s knowledge-source surface.
Three routes to the same 35 actions. The Workflow tab is generated from Chatbase's own manifest and carries its real ids, so it is copy-pasteable; the Code and CLI examples are the same call for any action on any app, so every app-specific value in them is a blank you fill in.
agent-auto-retrain-toggle Enable or disable automatic retraining every 7 days for an agent.
agent-clone Create a full deep-clone of an agent, including its sources (except Notion).
agent-create Create a new agent. Provide a URL to pre-train it from that website automatically.
agent-train Queue a training job for the agent. Asynchronous — poll Get Agent for `status`.
agent-update Partially update an agent's name and/or instructions. Only provided fields change.
conversation-get Get conversation metadata and its most recent messages. Only finds conversations created through the API — use the messages endpoint's cursor to page further back.
conversation-messages-list List a conversation's messages, paginating backward from the newest. Each page is chronological oldest to newest.
conversation-user-list List conversations for a specific user under an agent, ordered by last activity.
lead-list List leads captured by an agent's lead form. Uses the legacy v1 API — v2 has no leads endpoint as of this writing.
message-retry Retry generating an assistant response from a given message, truncating the conversation at that point.
source-create Create a text, Q&A, or link knowledge source for an agent. File uploads use a separate endpoint this app does not cover — see the README.
source-delete Mark a source for deletion. Restore it before this point with Restore Source.
source-summary-get Aggregated counts and sizes per source type, and whether a retrain is pending.
source-update Update a text, Q&A, or link source. Only send fields relevant to that source's own type — a link's URL is immutable; delete and recreate to change it.
ticket-list List helpdesk tickets for an agent, with filters and cursor pagination. Filters combine with AND.
ticket-message-add Post a customer-visible reply on a ticket. Markdown in, sanitized HTML out. A 201 confirms the reply was recorded, not delivered.
ticket-messages-list List a ticket's message thread. Defaults to chronological order, oldest first.
ticket-search Free-text search over ticket messages, ranked by relevance. Not paginated.
ticket-status-list List the active (non-archived) ticket statuses for an agent, ordered by category then position. Answers a bare array — no pagination.
tool-result-submit Submit the result of a client-side action the agent invoked. Continue the conversation afterward with Chat With Agent (message omitted).
whatsapp-template-list List approved WhatsApp templates across the agent's connected WhatsApp Business Accounts.
whatsapp-template-send Send an approved WhatsApp template to a phone number from one of the agent's connected numbers.
A workflow step names the app and the action, and the editor fills in the
connection when you pick one. This is the Step shape from the
workflow spec, carrying Chatbase's real ids.
{
"manifestVersion": "2",
"name": "chatbase-example",
"steps": [
{
"id": "agent-auto-retrain-toggle",
"uses": {
"app": "io.w6w.chatbase",
"action": "agent-auto-retrain-toggle",
"connection": "conn_YOUR_CONNECTION_ID"
},
"with": {
"agentId": "<agentId>",
"enabled": "<enabled>"
}
}
]
}agent-auto-retrain-toggle agent-chat agent-clone agent-create agent-get +30 more actions available
Every app-specific value here is a blank you have to fill in. An
app action is reached through the connection that authenticates it, so the
address is a connection id, not the app id — and connections belong to your account,
so a public page cannot know yours. Create one for Chatbase, then fill in
the three blanks: conn_YOUR_CONNECTION_ID, the action key, and the
parameters that action declares. The call itself is real — the shape is transcribed
from the studio's own snippet builder, which prints the same kind of blanks — but
nothing in it is specific to Chatbase. The Workflow tab is where this app's
real ids are.
npm install @w6w/sdkyarn add @w6w/sdkpnpm add @w6w/sdkdeno add npm:@w6w/sdkimport { W6wClient, isActionRun } from "@w6w/sdk";
// Reads W6W_BASE_URL and W6W_TOKEN from the environment when omitted.
const client = new W6wClient();
const envelope = await client.run({
urn: "conn_YOUR_CONNECTION_ID",
action: "agent-auto-retrain-toggle",
payload: {
agentId: "<value>",
enabled: "<value>",
},
});
if (isActionRun(envelope)) console.log(envelope.value); npm install -g @w6w/cli w6w run conn_YOUR_CONNECTION_ID --action agent-auto-retrain-toggle --payload '{"agentId":"<value>","enabled":"<value>"}' Give an AI agent Chatbase — without giving it Chatbase's credentials. One MCP endpoint exposes every app, function and workflow the caller is entitled to, as tools it can discover and run. Access is granted per team while we onboard.
One tool call{
"name": "w6w_invoke",
"arguments": {
"ref": "app:io.w6w.chatbase#agent-auto-retrain-toggle",
"input": {
"agentId": "<agentId>",
"enabled": "<enabled>"
}
}
}
Every tool names its target with a single ref. The
app: form above doesn't name a connection at all — the
host resolves which of the caller's Chatbase connections to sign
with, and refuses rather than guesses when the answer is ambiguous.
The token is attached host-side, at the moment of the call. It is never a tool argument, never in the model's context, and never in a transcript — so a prompt injection has nothing to exfiltrate.
Tools are derived per end user from what that person has actually connected and is entitled to — not one shared bot identity carrying the union of everyone's access.
Multi-step work runs on the workflow engine and returns a run handle the agent can poll — retries, branching and state survive the conversation that started them.
Chatbase's declared health checks are on the surface too, so an agent can tell "the vendor is down" from "your credential expired" before it burns a retry on either.
The MCP surface is part of the hosted platform. Chatbase itself is MIT, and the runtime that executes it is source-available (FSL).
Chatbase declares its own checks, so its health is a property of the app rather than something the host guesses at.
Chatbase publishes no usable status page (its Statuspage instance is an unclaimed decoy and status.chatbase.co is a dead, expired-certificate host) — this reads the API's own unauthenticated GET /health instead, the best signal Chatbase publishes for whether the platform is up.
Reads X-RateLimit-Limit / X-RateLimit-Remaining / X-RateLimit-Reset from GET /agents?limit=1 — present on every v2 response. Chatbase's other quota, message credits, has no readable balance endpoint and only surfaces as an error at chat time.