An AI agent in Helpdesky answers customers from your published articles and the pages you add, on your own AI key. You choose how much it does on its own, whether it reads like a colleague or is clearly an assistant, and how readily it hands off to a person.
Each agent is a distinct teammate with its own name, voice, skills and guardrails. Its mode decides how much it does on its own, and you can change it at any time.
Suggests a reply in the inbox. A teammate reviews, edits and sends it. Suggestions are instant and come with the agent's confidence and the sources it used; your team can use, edit or dismiss them.
Drafts a reply and waits for one-click approval before it goes out. Drafts appear in the inbox thread with Send, Edit and Discard. Nothing reaches the customer until a teammate approves it.
Replies on its own within the pacing and guardrails you set. Your team sees every reply in the inbox, can pause the agent on any conversation, and gets an email when the agent hands off.
Start with Copilot (a human sends every reply) and move an agent to Review or Autopilot once you trust it. Several agents can run side by side, each on its own channels.
Give each agent a name, a title and an avatar. Then decide how openly it shows what it is.
With the label shown, replies are marked as written by an AI and the agent can introduce itself as your team’s assistant in its first reply. With the label hidden, it reads like a colleague.
Hiding the AI label may be restricted where your customers live; compliance with disclosure laws is your responsibility. Whatever you choose, the agent admits it is an AI when asked. That is fixed in every preset.
Personality sets the AI label, the first-reply intro, the sign-off and how fast the agent replies, separately for live chat and for tickets & email. Pick a preset or set each value yourself.
Reads like a colleague: natural pauses that grow with reply length, no AI label, no bot intro, signs with its first name.
Clearly an assistant: AI label on, introduces itself in its first reply, offers a human when unsure, replies promptly.
Fastest possible: no pause on any channel, no typing bubble, AI label on.
Short messages, the customer is waiting. The reply delay is measured in seconds, and “like a busy human” grows with reply length.
Longer messages; replies are expected to take a while. The delay is measured in minutes, per channel: widget tickets, ticket center, contact form and email.
Pacing is enforced in Autopilot. Copilot suggestions are always instant, and Review drafts are prepared right away and wait for approval. Change any value by hand and the preset reads “Custom”; re-select a preset to restore it.
Carefulness bundles the guardrails into three presets. Each guardrail can also be set on its own.
Only answers with sources, needs high confidence, hands off to a human readily.
Answers when reasonably confident, hands off when the question is outside its knowledge.
Answers more often; hands off only when one of your hand-off rules applies.
Knowledge is what the agent knows; skills are how it acts on it. Instructions set the voice.
Your published help center articles plus the external URLs you add as data sources, found with semantic search over article embeddings. Agents never answer from the open web, and knowledge is managed once, in the help center, for every agent and for Ask AI.
Named instruction blocks any agent can pick up. Each skill has a one-line “use when” (“the customer asks for a refund or mentions a chargeback”) and the step-by-step instructions, policies and phrasing to follow.
Describe who the agent is, how it writes and what it should always or never do. Tone presets (friendly, professional, concise, warm) insert starter text; the sign-off can be none, the first name or custom text.
Add past replies from your team as style examples: they are shown to the model as reference answers so the agent matches your tone.
Nothing changes behind your back. Corrections are collected, turned into proposals, and applied only when you approve them. The style examples you add by hand stay as they are.
When a team member edits a Copilot suggestion or a Review draft before sending, or gives a reply a thumbs-down with a short reason, the correction is recorded against the agent.
Helpdesky turns those corrections into proposed changes to the agent’s skills or instructions: a new skill, an edit to a skill, or an edit to the instructions. Generated weekly, or on demand from the Suggestions tab.
Each suggestion shows the corrections behind it and a before/after diff. Approve it, edit and approve, or reject it. Analytics track the corrections and the suggestions your team used.
Agents work inside the same inbox as your team, so stepping in is one click away.
Chat with the agent as a customer, per channel, with your unsaved changes applied. See what it decided, its confidence, and which sources and skills it used. “Try on a recent conversation” runs the agent against the latest customer message of a real thread; nothing is sent or saved on the conversation.
Per agent, with a daily trend: replies sent, conversations handled, resolved without a human, hand-off rate, suggestions used and dismissed, thumbs up, average confidence and corrections. The Activity tab lists every run: playground, dry runs and inbox suggestions.
There is no AI surcharge and no Helpdesky AI quota. Agents use the provider key saved under Settings → AI Settings, and the provider bills the usage.
OpenAI, Anthropic, Google, xAI, DeepSeek, Qwen, Mistral and Perplexity, with a choice of model per provider. The same key powers Ask AI and the content help in the editor.
Stored encrypted for this workspace, visible only to the owner, and never shared across accounts. Remove it and every AI feature stops; agents can be set up before a key is added, but nothing runs until it is.
AI agents are included on every plan, Free included. You pay your provider for the tokens the agents use; the editor flags the places that spend them, such as automatic drafting and suggestion runs. Conversations past your plan’s monthly limit stay locked in the inbox and are left out of Copilot suggestions, dry runs and learning.
Ask AI is the self-serve layer: quick answers from your published articles, with links to the sources, in the chat widget and on the help center, before anyone writes in. It needs no agent and no conversation.
AI agents take over once a conversation exists: they draft or send replies inside the inbox, on live chat, tickets and email, with a name, a personality and guardrails. Both answer from the same help center and run on the same key. Try Ask AI on the homepage.
AI agents are new and being tested with early users. Everything on this page is in the product today; we are refining the editor, the presets and the learning loop as teams put agents to work. Start in Copilot, watch the Activity tab, and tell us what you would change.
That is your choice per agent. With the AI label shown (the Transparent and Instant presets), replies are marked as written by an AI and the agent introduces itself as your assistant. With the label hidden (Human-like), it reads like a colleague. In every case the agent admits it is an AI when asked, and disclosure laws where your customers live are your responsibility.
Only in Autopilot, and only within the guardrails you set: confidence threshold, required sources, hand-off rules, never-discuss topics and a maximum number of replies per conversation. Copilot and Review keep a human on every reply.
Your published help center articles and the external URLs you add as data sources, searched semantically. It does not browse the web. Skills add your policies and phrasing on top.
Below the confidence threshold it hands off to your team (or stays quiet when hand-off is “only on rules”), the conversation shows up under “Needs human”, and your team can get an email. The hand-off reason is shown on the conversation.
Every edit your team makes before sending, and every thumbs-down, is recorded as a correction. Helpdesky turns corrections into suggested changes to skills or instructions; you approve, edit or reject each one. You can also add your best past replies as examples.
AI agents are included on every plan. You bring your own provider key, and the provider bills the tokens used. See pricing for plan limits.
No. Those are for your coding agent (Replit, Cursor, Claude Code and friends) to set Helpdesky up and write articles for you. AI agents on this page are support teammates that talk to your customers. See AI Skill and MCP server for the coding-agent story.
Create an agent in Copilot mode, add your AI key, and watch it suggest replies from your own articles. No credit card required.
Start freeAlready on Helpdesky? Open AI Agents in the dashboard.
