Start from your own interface.
Use a custom launcher, prepare an opening message, or reopen an existing conversation.
Bring support into your app, identify the customer behind each conversation, and give agents access to the tools behind the work.
Open source · Self-host free, or let us run it
Your app starts the conversation. Agents work with the tools you connect.
Install Helpin, check the services, and inspect logs from the terminal.
Explore the CLISDKsAdd support chat to your app and identify the customer behind each conversation.
Explore the SDKsMCP connectionsConnect AI clients and integrations to Helpin, or give Helpin agents tools from another system.
Explore MCPEvents & automationStart agent work from GitHub and GitLab events, workspace changes, or a schedule.
Explore events and automationStart a conversation from your own interface. Associate it with the signed-in customer and their company, then keep the exchange in Helpin.
npm install @helpin-ai/[email protected]import { helpinClient } from '@helpin-ai/sdk-js';
const client = helpinClient({
widgetKey: 'YOUR_PUBLIC_WIDGET_KEY',
host: 'https://client.helpin.ai',
autoBoot: false,
});
client?.openNewMessage(
'My export is missing contacts. Can you help?'
);
Use your widget’s public key and host. For self-hosting, set host to your installation’s public widget URL.
Use a custom launcher, prepare an opening message, or reopen an existing conversation.
Identify signed-in customers so the conversation has a clear account context.
Bring an article into the widget from your product’s help menu or interface.
Verify the person behind the conversation, then let agents investigate through selected tools. Customer identity and permission to access data are separate controls.
Your backend signs an identity proof with HMAC-SHA256. Helpin validates the proof when the widget identifies the customer. Keep the signing secret on your server, and turn on “Require server-signed identities” to reject unsigned ones. Set up customer identity →
No. You choose the agent’s tools and approval rules separately. Your connected tool server must also enforce which customer’s data it can return. Configure tool access →
The agent gets what it needs to investigate—not unrestricted access.
AI clients and integrations work with Helpin through OAuth or a scoped service token. Helpin agents can use tools from other MCP servers.
Let an authorized AI client or integration search your docs and use the workspace capabilities you grant. A workspace manager turns it on.
Connect an external MCP server and select the tools each agent can use, with approvals where required. The deployment must enable external servers first.
Connect GitHub and GitLab events, workspace changes, and schedules to agent workflows. A matching rule starts the job with the agent’s selected tools, instructions, and approvals.
Inspect trigger activity, including skipped events, separately from agent runs.
Use an SDK to put support inside your product. Use MCP to connect AI clients and integrations to your workspace, or to give Helpin agents tools from other systems.
@helpin-ai/sdk-js for any web app, @helpin-ai/react for React, @helpin-ai/nextjs for Next.js, and @helpin-ai/vue for Vue 3 and Nuxt. Install the framework package alongside @helpin-ai/sdk-js.
Yes. Set host to your installation’s public widget URL and add your site to the permitted website origins.
No. The widget key is public and only identifies your widget. Integrations reach your workspace through MCP, with OAuth or a scoped service token, and identity signing uses a separate server secret.
Your backend signs an identity proof with HMAC-SHA256, and Helpin checks it when the widget identifies the customer. Turn on “Require server-signed identities” to reject unsigned identities; until then, they’re accepted as unverified. Verification doesn’t grant access to connected tools.
Public MCP lets an external AI client or integration work with Helpin. External MCP lets a Helpin agent use tools from another system.
Both connections are in controlled beta. A workspace manager turns on public MCP; external MCP servers must first be enabled for the deployment. What each connection can do depends on granted scopes, user permissions, and enabled modules.
GitHub and GitLab repository events, such as a pull request opened, merged, or closed, or a review requested, plus workspace events and schedules. A matching automation rule starts the selected agent with its tools and approvals. Outbound webhooks aren’t available yet.
Start a free Cloud trial, or self-host the open-source edition and build against your own installation.
Open source · Self-host free, or let us run it