Open-source IT helpdesk agent

Answers your team’s IT questions in chat, from your own docs.

Deskhand lives in Slack, Microsoft Teams, and Google Chat. It searches your how-to docs, cites them, and checks with the person asking before it changes anything. You host it.

Get started View on GitHub

# it-helpThread

Sam

@Deskhand How do I connect to the VPN?
Deskhand App

Open GlobalProtect, enter vpn.acme.com as the portal, and sign in with your work account. Approve the Okta push when it appears.

Source: VPN setup (Mac and Windows)
Sam

It says "gateway unreachable". Can you open a ticket?
Deskhand App

Approval needed: Create a Jira issue in IT: "VPN gateway unreachable"

jira_create_issue

Only Sam can approve or deny.

Works where your team chats

DM it or mention it in a channel in Slack, Microsoft Teams, or Google Chat. In channels and spaces it answers in the thread. Slack connects over Socket Mode, so you need no public URL to start.

Answers from your docs

Point it at a Google Drive folder, a Confluence space or page tree, or a folder of Markdown. It searches them, answers, and links the page it used.

Extensible, with safe writes

Add tools with an MCP server in config.yaml or a TypeScript extension in your project. Anything that writes, like opening a Jira ticket, waits until the person who asked confirms it.

Start from the template

Clone the starter project; the engine comes from the deskhand npm package. Put your model key and Slack tokens in .env, settings in config/config.yaml, and your how-to docs in docs/, then start it with Docker Compose. Upgrade later with npm install deskhand@latest.

Read the quickstart

git clone https://github.com/OWNER/deskhand-starter my-it-agent
cd my-it-agent
npm install
cp .env.example .env
cp -r /path/to/your/runbooks/. docs/
docker compose -f deploy/docker-compose.yml up

IT Housecall

When chat isn’t enough, hand off to a live screen share with IT Housecall.

Some problems need eyes on the screen. Turn on the handoff and Deskhand offers a help link when the docs don’t solve it. The person presses one button, shares their screen, and talks to an AI that sees it and walks them through the fix.