Product

AI agent and Copilot

Configure the AI agent's mode, triage, procedures and tools, and use Copilot to help agents reply faster.

AI works in two ways in ClickFix:

AI agent: triages customer conversations, drafts replies grounded in your knowledge base, or sends the reply itself under the conditions you choose.
Copilot: works next to the agent in the inbox; suggests replies, summarizes and edits text.

AI usage is charged against the AI credits in your plan. When credits run out, the agent hands conversations over to a human.

Configure the agent under Organization → AI. The page has four tabs: Agent, Procedures, Tools and Run log.

Mode

Mode	What it does
Off	The agent writes no replies. With triage on, only topic, priority and a summary are extracted.
Summary only	Every new message is triaged and summarized for agents; nothing is sent to the customer.
Draft + approval	The agent drafts a reply grounded in the knowledge base; an agent approves and sends it.
Autonomous	Sends the reply itself when it passes the confidence threshold and can cite a source; otherwise hands over to a human with a summary. Stays silent once a human has replied.

Mode changes apply immediately. If you're new, we recommend starting with Draft + approval and moving to autonomous once you're happy with the reply quality.

Behavior

Automatic triage: extracts the language, intent, category, sentiment and urgency of incoming messages.
Persona and tone: your brand voice and style preferences. This text can't override safety rules.
Confidence threshold: in autonomous mode a reply is sent only when the model's confidence is above this value; otherwise it waits as a draft and is handed to a human.
Handoff keywords: if a customer message contains one of these words (e.g. "agent"), the agent hands over to a human right away.

Scope and working hours

Channels: the agent only works on the selected channel types. With none selected it never steps in.
Categories: when set, the agent only replies autonomously in conversations triaged into one of these categories.
Reply language, Time zone, Days and Working hours.

You can also switch the agent on or off per channel with the AI agent option in each channel's Settings.

Knowledge source

The agent grounds customer-facing replies in the public, published articles of your knowledge base. Write your frequent questions as articles for better answers. See Knowledge base.

A procedure is the steps the agent follows for a specific customer intent. For a refund request, for example: ask for the order number first, call the order tool, hand over to a human if more than 14 days have passed.

Start with Procedures → New procedure.
Enter the Intent exactly as triage produces it (e.g. refund_request) and pick the language. Intent and language can't be changed later.
Write the title and steps, save, then publish.

Only the published version is used. A procedure can't grant the agent new tools and can't override safety rules.

Tools

Built-in tools come with every organization and only work inside the related conversation:

Tool	What it does
`lookup_contact`	Reads the contact's details in the conversation
`conversation_history`	Reads the conversation's public message history
`knowledge_search`	Searches published knowledge base articles
`create_note`	Adds an internal note (with agent approval)
`set_status`	Changes the conversation status (with agent approval)

Organization tools (HTTP) turn an endpoint in your own system into a tool the agent can call (order, license or invoice lookup, for example):

The agent calls the tool with a JSON POST request.
The Tool name starts with a lowercase letter and is 3-50 characters (a-z, 0-9, _). It can't be changed later.
The Daily call limit is between 1 and 1000.
With Needs approval on, the agent only proposes the call and an agent approves it. Keep it on for tools that change data.
Credentials are stored encrypted and never shown again. Tools aren't deleted; they're disabled.

Test chat and run log

In the Test chat, write a message as a customer would and see the agent's reply and the sources it used. Every test is a real run and uses credits.

The Run log records every AI run with personal data masked: input, knowledge chunks, tool calls, output, tokens, cost and latency. Use it to see why a reply was given.

Safety

The agent stops requests that look like prompt injection or would produce unsafe output.
When daily call, concurrency or budget limits are reached, no new run starts.
Set how long AI logs are kept under Privacy & GDPR → Retention. See KVKK tools.

Copilot

With a conversation open in the inbox, Copilot can:

Suggest a reply and Summarize conversation
Rewrite, Translate, Change tone, Shorten or Expand your draft
Suggest next step
Ask Copilot: a free-form question about the conversation

Open the Copilot menu from the reply editor or with keyboard shortcuts (? lists them all). Copilot suggestions never reach the customer unless you send them.

AI during set-up

When you open a new organization, the set-up wizard can use your website and a few answers to suggest teams, SLAs, canned replies, starter articles and the agent's persona. See Getting started.