AI Assistant
Let the AI Assistant answer customers or draft replies.
Enable or disable the ai assistant for the company
An Admin turns the AI Assistant module on or off for the whole company — a master switch independent of the operating mode and of any per-channel enablement.
Select the ai operating mode
An Admin chooses what the AI Assistant does company-wide — Copilot (the AI only drafts for operators), Auto Reply (the AI also answers customers directly) or neither (Disabled) — and, per connected channel, whether the assistant is enabled on that channel at all. The modes build on each other, so the screen shows them as two switches, the second appearing once the first is on.
Configure the ai's operating instructions
An Admin writes free-form instructions that steer how the AI Assistant behaves — what to emphasize, what to always mention, how to handle specific situations — supplied on every request alongside the persona and the retrieved sources.
Manage the ai's sources
An Admin uploads documents and writes text or FAQ entries that ground the AI's answers, activates and deactivates them, and follows their indexing state.
Configure the ai's persona and tone of voice
An Admin describes the identity the AI presents (e.g. a name, a role framing) and the tone it should use, supplied on every request.
Configure handover rules
An Admin defines categories and conditions that always transfer a conversation to a human, regardless of groundedness.
Configure whether the ai may answer beyond its sources
An Admin sets the company's groundedness policy: whether the AI must hand over any reply not supported by the company's own configured sources, or may also answer from its general knowledge.
Configure forbidden topics
An Admin lists topics the AI must never answer on or make commitments about — the AI hands over instead, independent of groundedness.
Test the ai's configured behavior
An Admin exchanges test messages with the AI in a sandbox that runs the full configured pipeline — instructions, persona, sources, groundedness, forbidden topics, handover rules — without ever reaching a real customer or any channel.
Configure the fallback message
An Admin writes the message sent to the customer, per language, when the AI cannot produce a reply.
Configure customer disclosure of ai replies
An Admin edits, per language, the disclosure text that tells customers an Auto Reply message comes from the AI. Disclosure is always on.
Answer a customer automatically in auto reply mode
An inbound customer message on a channel where Auto Reply is active is answered by the AI Assistant without operator involvement — the AI becomes the conversation's responsible party — or the customer receives the fallback message and the conversation transfers to an operator.
Assemble the context supplied to the ai
Before every call to the AI/LLM Service — Auto Reply or Copilot — the system assembles a bounded request context: the current message, recent conversation history, permitted customer-profile fields, retrieved source passages, configured rules, channel restrictions and any channel-specific reply context, all limited to the requesting your visibility scope and never crossing tenants.
Reply in the customer's language
The AI Assistant replies in the language of the customer's incoming message, even when it differs from the customer's stored profile language.
Generate a reply suggestion for the operator
An operator asks the AI Copilot to draft a reply to the current conversation; the suggestion is shown in the composer for the operator to accept, edit or dismiss — nothing is sent automatically.
Regenerate an alternative suggestion
An operator who is not satisfied with a generated suggestion asks for a different one.
Improve or rephrase a drafted reply
An operator who has written their own draft asks the AI to improve or rephrase it, rather than generate a reply from scratch.
Edit an ai suggestion and send it
An operator changes the wording of a generated suggestion before sending it.
Accept an ai suggestion and send it unchanged
An operator sends a generated suggestion exactly as drafted.
Dismiss an ai suggestion
An operator discards a generated suggestion without sending it.
Ask to be transferred to a human operator
A customer being answered by the AI in Auto Reply mode asks to speak with a person; the AI recognizes the request and the conversation transfers immediately, with its complete history.
Transfer for lack of groundedness
When the AI's drafted Auto Reply is not supported by the company's configured sources and , the conversation transfers to an operator instead of sending the ungrounded reply.
Transfer by configured rule or category
When the customer's message or the AI's intended reply matches a configured handover rule or a forbidden topic, the conversation transfers to an operator instead of being answered automatically — independent of groundedness.
Hand over the complete ai conversation history
When a conversation transfers from the AI to a human, the receiving operator sees the whole conversation exactly as it happened — nothing summarized, filtered or hidden.
Stop automatic ai replies after takeover
Once a human operator has explicitly taken over an AI-held conversation, the AI never auto-sends another reply to it.
Send the fallback message and transfer when the ai cannot reply
When the AI/LLM Service times out, errors, or returns an empty result, the customer receives the configured fallback message and the conversation transfers to an operator — never a silent failure.
Disclose that a reply comes from an ai
Every Auto Reply message sent to a customer carries the configured disclosure text.
Record the provenance of every message
Every message in a conversation records who produced it — Operator, AI, Automation or System — and, for a human sender, the real author behind the company's shared channel identity.
Record that an operator-sent message was ai-drafted
When an operator sends a Copilot suggestion — unchanged or edited — the message carries Operator provenance, plus a separate flag recording that the AI drafted it.
Retain generated suggestions for staff only
Every generated Copilot suggestion — sent, edited, or dismissed — is kept with the conversation for staff to review, and is never visible to the customer.
Keep ai messages within channel restrictions
An AI-originated message — Auto Reply or a sent Copilot suggestion — obeys the same channel rules as any other outbound message: messaging window, required templates, and supported content types.
Set the provider commission and ai plan limits
The Platform Operator sets the commission added to the AI provider's cost, per platform and per company, and sets what a company's plan includes: the AI Assistant feature, a monthly spend cap and the highest model tier.
View ai usage and spend
An Admin reviews what the AI has cost the company this month and in previous months, by pool, channel, operator and tier, and how much remains; an operator sees their own Copilot usage and what remains.
Set the auto reply and company ai budgets
An Admin limits the monthly AI spend of the whole company, of Auto Reply, and of each channel connection, within the plan cap.
Set copilot budgets and model tiers per role and operator
An Admin sets how much Copilot each operator can use per month and which quality tier serves them, as a default per role with a per-user override.
Reach an ai usage limit
When a monthly AI limit is reached, the AI stops at that layer in a defined way: Auto Reply hands conversations to humans, Copilot is disabled for the affected operator, indexing waits and the sandbox is refused. No customer is left without an answer.
Set up the ai assistant from a preset
An Admin follows the guided setup: picks a use-case preset and a strictness, adds sources, tests, sets budgets, reviews the go-live checklist and starts in Copilot mode or on one channel in Auto Reply.