Agents and AI Agents

The Agents page is the account directory for the people and the autonomous AI profiles that work on customer conversations in Teloring.

Use it to:

• Invite human team members and manage their access.
• Choose each person's role, language, login security, and voice access.
• Let people reset their own password or move their own sign-in email — safely, through their inbox.
• Resend invitations, deactivate access, or delete a profile.
• Create AI Agents and control how they speak, what they know, what they collect, and how they hand a conversation back to a human.

Open Admin → Agents, Teams & Roles → Agents from the Teloring sidebar.

Grouping agents

Once your agents exist, group them into Teams — Sales, Support, Billing — so a conversation can be routed to a whole team instead of one person. Teams is the second page under Agents, Teams & Roles.

The Department field on an agent profile is not a team. It is a free-text label for display and search only.

Account isolation

Everything on this page belongs only to the current Teloring account. An agent or AI Agent created here is never visible in another account.

Human agents and AI Agents at a glance

Both types live in the same table, but they work very differently.

	Human agent	AI Agent
What it is	A person who signs in to Teloring.	An autonomous conversation profile. No person behind it.
Sign-in	Email + password, with optional email 2FA.	None. No email, no password.
What you configure	Identity, role, language, 2FA, department, notes, voice.	Identity, writing style, knowledge, rules, goals, outcomes, working hours.
Conversation assignment	Picked in the Assigned to control.	Same control, marked with a robot icon.
Once assigned	The person answers in the workspace.	Teloring opens an AI session and the AI answers incoming messages.
Turn it off	Deactivate (reversible).	Set the profile Status to Inactive.
Delete it	Yes, an admin can delete the profile.	Not from this page — no delete action exists yet.
Can join a Team	Yes	Yes
Counts as "online"	Yes, while signed in to Teloring.	Always online. An AI Agent is software, so it never goes offline.

Two different meanings of "online"

This distinction matters once you use Teams, and it is deliberate:

• Team auto-assignment treats an AI Agent as always online, so an AI member can pick up work at 3 a.m. Use a team's Only real agents rule when you do not want that.
• Presence features — the Team Chat green dot, the live-chat "agents online" count, and Studio's Agent availability condition — ask "is a person there?" and never count AI Agents.

Who can do what

Any signed-in member can open the page and read the directory. Everything that changes something is enforced on the server.

Action	Permission it needs
View the directory and search it	Agents → Read
Create a human agent or an AI Agent	Agents → Create
Edit an agent or an AI Agent profile	Agents → Update
Change per-agent voice access	Agents → Update
Resend an invitation	Agents → Update
Activate or deactivate a human agent	Agents → Update
Delete an agent profile	Agents → Delete
Change account voice settings	My Ring → Update
Reset their own password	None — always allowed
Reset somebody else's password	Agents → Update
Change their own sign-in email	None — always allowed
Change somebody else's sign-in email	Agents → Update
Assign a conversation to a human or AI Agent	Can assign conversations on that inbox
Create, edit, or delete a Team	Teams → Create / Update / Delete

See Roles and Permissions for how these are granted.

Two rules apply to everyone, whatever their role:

• You cannot deactivate or delete your own profile here.
• Nobody — not even an admin — types a password or a new email address on someone else's behalf. Both always travel through a one-time link sent to the agent's own inbox.

Page overview

Element	What it does
Add AI Agent	Asks for a name, then creates the AI profile and its directory row.
Add Agent	Opens the human-agent invitation form.
Total Agents	Every row in the account: active people, pending invitations, inactive profiles, and AI Agents.
Active Agents	Rows whose internal active flag is on. This includes pending humans and active AI Agents, so it can be higher than the number of green Active labels. Read the Status column for the exact onboarding state.
Voice channel	Account-wide switch for Teloring browser calling. It must be on before voice can be enabled for any individual person.
Search	Filters the loaded table by name, email, phone, or department. It does not search role, status, 2FA, or creation date.
Agents table	The directory and its per-row actions.

Changes are account-wide, and the table reloads after each successful operation.

Understand the table

Column	Meaning
Name	Display name, optional description, a You badge on your own row, and an AI badge on AI rows. The colored circle is the first letter, tinted from the name.
Email	Sign-in address. AI rows show No login — AI agent.
Phone	Optional, informational only on this page.
Department	Free-text label such as Support or Sales. Display and search only — not a Team. AI Agents are always filed under AI.
Role	The role this person holds — one of your account's own roles. AI rows show AI Agent, which is a label, not a role.
2FA	Whether that person's own 2FA switch is on. AI rows show a dash.
Status	Active, Pending, or Inactive.
Created	Creation date and time, in your browser's format.
Actions	See Row actions.

What each status means

Status	Meaning	Can they sign in or be assigned work?
Active human	Registered and enabled.	Yes.
Pending human	Invited, but has not chosen a password yet.	Sign-in is blocked. The profile can still appear in assignment pickers, so avoid assigning work to a Pending row.
Inactive human	An admin deactivated the profile.	New sign-ins are blocked.
Active AI Agent	Ready to be assigned.	Yes.
Inactive AI Agent	Profile status is disabled.	New assignments are rejected.

Deactivating does not kick someone out

Deactivation blocks new sign-ins, but it does not end a browser session that is already open. When access must stop immediately, also revoke that person's sessions in Settings → Security & Login → Active sessions. (Deleting a profile does revoke sessions.)

Row actions

The icons at the end of each row change with the row's type, its status, and your role.

Icon	Appears on	What it does
Edit (pencil)	Every row	Human → the agent form. AI → the AI profile editor. Saving needs Agents → Update.
Resend confirmation email (envelope)	Agents → Update, on a Pending person	Sends a fresh 1-hour invitation and stops the previous invitation link from working.
Reset password (padlock)	Your own row always; anyone else's with Agents → Update	Emails a one-time link so a new password can be chosen. See Passwords.
Change email (envelope with arrow)	Your own row always; anyone else's with Agents → Update	Starts a sign-in-address change, confirmed from the current inbox. See Change a sign-in email.
Deactivate / Activate	Agents → Update, on someone else	Turns the profile off or back on. Reversible.
Delete agent (bin)	Agents → Delete, on someone else	Permanently removes the profile after typing DELETE.

Reset password and Change email are the two actions that never need a permission — everybody gets them for their own row. Extending either to somebody else is Agents → Update.

note

Password reset does not appear on a Pending row — there is no password to reset yet. Use Resend confirmation email instead.

---

Human agents

Invite a person

1. Open Admin → Agents.
2. Click Add Agent.
3. Fill in the name and email (both required).
4. Choose the role and default language.
5. Optionally add phone, department, description, 2FA, and comments.
6. Click Create Agent.

Teloring creates a Pending profile and emails a one-time password-setup link.

• The link is valid for 1 hour and can be used once.
• Until the person chooses a password their row stays Pending and sign-in is blocked.
• No password is set from this form. Nobody in the account ever sees or chooses another person's password.
• Sending a new invitation cancels the person's earlier unused invitation links.
• If the email fails to send, the profile is still created. Fix the address or the mail problem and use Resend confirmation email.

Field reference

Field	Required	What it controls
Agent Name	Yes	The name shown across Teloring: assignments, messages, audit entries, Team Chat.
Email	Yes	The sign-in address. It must be unique across all of Teloring. In Edit it is read-only — move it with Change email.
Phone	No	Internal contact number. Local or international format.
Role	No	One of your account's roles. The list is your own, and the role's description appears underneath so you can confirm the choice. Defaults to the account's Agent role.
Department	No	Free-text label, used for display and search only. It does not create a team, a queue, or a permission group. To group agents for routing, use Teams.
Default Language	No	English or Hebrew. Sets the person's workspace language and the language of their invitation, reset, and 2FA emails.
Description	No	One line about the person's responsibility, shown under their name.
Two-Factor Authentication	No	When on, sign-in also requires an emailed code.
2FA Method	No	Leave on Email. See What is not finished yet.
Comments	No	Internal notes. Never shown to customers, but any member who can open the profile can read them.
Voice for this agent	Edit only	Turns on this person's browser softphone, once account voice is on.

Choosing a role

The Role list contains your account's own roles. Every account starts with five — Owner, Team Leader, Marketing, Agent and Viewer — and you can rename, re-scope, duplicate or delete any of them except Owner.

Picking a role decides everything the person can reach: which pages, which inboxes, and what they may do in each. Read the description under the picker if you are unsure, or open Admin → Agents, Teams & Roles → Roles & Permissions to see the full grid.

If the person is…	Start with
A front-line agent answering customers	Agent
A support manager who is not the account owner	Team Leader
Working on campaigns, automations and reports	Marketing
A stakeholder who only wants dashboards	Viewer
A co-owner of the workspace	Owner
Anything else	Duplicate the closest role and adjust it

See Roles and Permissions.

Owner is full access

Owner can reach everything, including billing and account deletion, and its permissions cannot be narrowed. Give it only to people who genuinely own the workspace — but give it to at least two of them, so nobody is ever locked out.

Edit a person

Click the pencil on a human row. You can change name, phone, role, department, default language, description, 2FA, comments, and voice access.

The email field is read-only here on purpose — see below.

Passwords

Passwords are only ever chosen by the person themselves, from a one-time emailed link. There are three ways one gets issued.

Path	Who starts it	Effect on the current password
Invitation (new profile / resend)	Admin	No password exists yet.
Reset password — my own row	Anyone, for themselves	Keeps working until the new one is chosen, so a lost or slow email cannot lock you out.
Reset password — someone else	Admin	Stops working immediately. Use this when credentials may be compromised.
Forgot your password? on the sign-in page	Anyone, without signing in	Same as a self-reset. The page answers identically whether or not the address exists, so it cannot be used to discover who has an account.

Every one of those links:

• lasts 1 hour and works once;
• is stored only as a hash — the real value exists only inside the email;
• cancels the person's other unused links of the same kind (so a new password link no longer silently cancels a pending email-change confirmation);
• stops working if the profile is deactivated or its address changes in the meantime.

Password reset requests are rate limited per IP, and repeated requests for the same address are throttled, so one inbox cannot be flooded.

Password rules

A new password must:

• be at least 8 characters (and at most 72 bytes);
• contain an uppercase letter;
• contain a lowercase letter;
• contain a digit;
• contain one special character — anything visible that is not a letter or a digit. Spaces do not count.

The set-password page lists the same rules and checks them as you type.

Change a sign-in email

The sign-in address is an identity, so moving it is deliberately a two-step, inbox-confirmed operation.

1. Open the person's row (or the Edit form) and click Change email.
2. Confirm. Teloring emails a confirmation link to the address currently on file — never to a new one.
3. The recipient opens the link and types the new address there, once.
4. Only then does the sign-in address move. Teloring also notifies the old address that it is no longer the sign-in email.

Why it works this way: control of the existing inbox is what authorises the move. An admin — or someone who hijacked a dashboard session — cannot take over an agent's identity without being able to read the inbox that agent already uses.

The link lasts 1 hour, is single use, and nothing changes until it is opened.

Two-factor authentication

Per-person 2FA is separate from the account-wide Enforce 2FA for all members switch in Settings.

A code is required when either is true:

• the person's own Two-Factor Authentication switch is on;
• the account-wide enforcement setting is on.

The code is emailed, expires after 5 minutes, is stored only as a hash, and is attempt-throttled.

Teloring can also ask for a code when someone signs in from a browser it has not seen before, depending on your account's new-device setting.

note

Keep the method on Email. The WhatsApp option can be saved, but every code is still delivered by email.

Voice

Voice works at two levels, and the account level comes first.

1. Account voice

The Voice channel card switches Teloring browser calling on or off for the whole account.

• Needs My Ring → Update.
• It must be on before any individual can get voice credentials.
• Turning it off stops account voice sessions from starting.
• The card is a single switch. The account's concurrent-call limit exists in the backend but is not editable here.

2. Per-person voice

Open a person's profile and use Voice for this agent. When you enable it, Teloring:

• generates a private SIP extension and password;
• encrypts the password at rest;
• provisions the extension on the Teloring voice server;
• allows that person's call window to start voice sessions.

If provisioning fails, the operation reports an error and the switch reverts.

Reset voice credentials rotates the SIP password, ends the current voice session, and updates the provisioned extension. Use it if credentials may have leaked or the softphone can no longer register.

note

These controls are for human browser softphones. AI Agents never receive SIP credentials.

Deactivate or delete?

Choose	When	What survives
Deactivate	Leave of absence, suspension, or a reversible offboarding step.	Everything. The profile can be switched back on.
Delete agent	Permanent removal, after access and ownership have been reviewed.	Conversations, messages, audit entries, and names already copied into records.

Deleting a profile also revokes that person's dashboard sessions, outstanding password-setup links, trusted devices, and personal API tokens.

Before you delete:

1. Reassign any open conversations they still own.
2. Check Studio flows or processes that name that agent.
3. Turn voice off first — deleting the profile does not deprovision the voice extension.
4. Accept that history stays, without a live profile behind it.

---

AI Agents

An AI Agent talks to the customer directly. Do not confuse it with AI Copilot:

• AI Agent owns the conversation and answers the customer.
• AI Copilot only advises a human in the right-hand panel.
• They never run on the same conversation at the same time.

Create an AI Agent

1. Open Admin → Agents.
2. Click Add AI Agent.
3. Type a name when prompted, and confirm.
4. Find the new row and click its pencil to open the profile editor.
5. Complete the profile before you point real customers at it.

Creating the row does not attach it to any inbox or conversation.

Put an AI Agent to work

1. Open an active conversation.
2. Open Assigned to.
3. Pick the AI Agent (robot icon).
4. Teloring starts an AI session from a snapshot of the profile.
5. The conversation moves to the AI Agent queue, unless a Studio flow still owns it.
6. The AI sends its greeting when appropriate and answers incoming messages from then on.

Because each session uses a snapshot, editing a profile changes future sessions only. A conversation already running keeps the settings it started with.

warning

Setting a profile to Inactive blocks new assignments but does not stop a session that is already running. To stop one immediately, reassign that conversation to a human or back to the waiting line.

The profile editor

Eight tabs, one Save AI Profile button that saves all of them together.

Identity

Field	What it controls
Name	The sender name customers see, and the directory name.
Job title	Goes into the AI's instructions, e.g. Customer Care Specialist.
Gender	Tells the AI which self-reference forms to use in gendered languages.
Status	Active allows new assignments; Inactive blocks them. Saving also updates the directory row.
About / Bio	Explains the AI's role and the business it represents. Part of the instructions.
Internal note	For your team only. Never sent to the model or the customer.

Use the bio to describe responsibility and context — not to re-type a whole Knowledge Base.

Voice & Language

“Voice” here means writing style. It has nothing to do with phone calls.

Field	What it controls
Tone preset	Friendly, professional, casual, formal, playful, empathetic, or concise.
Response length	Short, medium, or long guidance — not a hard character limit.
Allowed languages	Which languages the AI should answer in. Hold Ctrl (Windows) or Command (macOS) to pick several. Select none to let it follow the customer.
AI disclosure	Always, If asked, or Never. “Never” tells the AI not to describe itself as automated — check that against the rules that apply to your business.
Signature	Appended to AI messages when not already present.
Greeting message	The first message, sent on assignment, if the conversation has not already been greeted by an AI.

Greeting variables:

Variable	Becomes
{{customer_name}}	The customer or contact name, or “there” if unknown.
{{agent_name}}	The AI Agent's name.
{{conversation_id}}	The current conversation ID.

note

Hebrew, Arabic, and English are the languages Teloring can reliably detect server-side. Russian, Spanish, and French shape the prompt but are not enforced.

Knowledge

Field	What it controls
Knowledge base	Which account Knowledge Base is searched before every AI turn.
Grounding mode	How strictly the AI must stay inside that Knowledge Base.
Confidence threshold	How similar a passage must be to be retrieved. Low is broadest, High is strictest.
Show citations	Asks the model to cite grounded information. Guidance, not a guaranteed format.

Grounding mode	Use it for
Strict KB only	Regulated or tightly controlled answers. A Knowledge Base must be selected before you can save.
KB + general language	The Knowledge Base is the source of truth, but the AI may phrase things naturally. A good default.
Open assistant	Broader answers, still preferring Knowledge Base context when it exists.

Build and test the Knowledge Base before switching to strict mode.

Rules

Field	What it controls
Forbidden topics	Comma-separated topics the AI must not discuss. If a generated answer contains one of these exact phrases, Teloring replaces it with a refusal/handoff sentence.
Do not promise	Comma-separated commitments the AI must not make, such as unapproved discounts or delivery dates.
Behavioral instructions	Free-text operating rules. Keep them short and non-contradictory.
Max turns	Enforced. Maximum customer turns before the session ends on a limit.
Max duration minutes	Enforced. Maximum session age, checked when a message is processed.
Max consecutive AI messages	Guidance only — see What is not finished yet.

Put factual material in the Knowledge Base, collection requirements in Goal, and only true behavioural rules here.

warning

The forbidden-topic filter replaces an unsafe answer, but replacing text does not move the conversation. Keep the Forbidden topic escalation reason enabled with a real routing action.

Goal

Goal decides whether the AI only answers, or also gathers information into your CRM.

Mode	Meaning	Use it?
Give info	Answer and help. No proactive collection.	Yes — for information and support profiles.
Get info	Intended for collection only.	No. It does not trigger collection (see below).
Both	Answer the customer and collect one missing required field at a time.	Yes — this is the working path for any collection.

Choose “Both” for collection

Get info does not reach the runtime's collection mode, so a profile set to it never asks for your fields. Use Both whenever the AI must collect data.

Add a field to collect

1. Set the mode to Both.
2. Click Add Field.
3. Drag a CRM field from the right-hand picker onto the card's Target field box. The target box is read-only — dragging is how it gets filled, and it also fills the label, internal slug, data type, and choice values.
4. Adjust the customer-facing label if you want.
5. Check the data type.
6. Optionally write a sample question.
7. Set the toggles.
8. Repeat for each field.

Property	Meaning
Label	The human name, used to phrase a natural question.
Data type	Text, number, email, phone, date, single select, multi select, boolean, or file.
Target field	Where the answer is saved in the CRM.
Sample question	Optional. Leave it empty and the AI writes its own question from the label and type.
Allowed values	For select types, the accepted values.
Mandatory	Counts towards “all mandatory fields collected”.
Re-ask if known	Whether the AI may ask again when the value already seems known. Guidance.
Overwrite if present	Off = never replace a non-empty CRM value. On = a new answer may replace it.

Supported destinations:

• customer.<field>
• contact.<field>
• conversation.custom_fields.<field>

Custom object targets do not save

The picker also lists your custom objects (deals, tasks, and so on) as object.<schema>.<field>. The AI runtime does not write those. Stick to Customer, Contact, and Conversation targets.

Answers are validated (email, phone, number, boolean, and choice values) before saving, and customer/conversation writes fire the usual Studio change events.

Conversation targets are custom fields

The three Conversation targets write under conversation.custom_fields. The Conversation priority target, for example, does not change the real priority control in the conversation header.

Two picker quirks to work around

• Type the label after dragging the target. Typing first can freeze the hidden slug after one character.
• With several field cards open, drag the CRM field onto the exact card you mean; a click can land on the first card instead.

On Success

A session can end successfully when the customer confirms it is resolved, or when every mandatory field has been collected and the customer then goes quiet until the follow-up timer.

Setting	Meaning
All mandatory fields collected	Lets completed collection become a quiet success after the follow-up timeout.
Customer confirmed resolved	Lets a clear “all good” or “thanks” close the session.
Closing message sample	Guidance for tone and intent. The real closing text is generated from the conversation and the customer's language.
Action	What happens after the closing message.

Action	Result
End and resolve	Resolves the conversation and clears the assignment.
No action	Ends the AI session and changes nothing else. Be careful: the conversation can stay assigned to the AI with no live session.
Send back to waiting line	Clears the assignment so a human can pick it up.
Assign to human agent	Assigns the chosen active person. If that person is no longer valid, it goes to the waiting line instead.

On Fail

The customer went quiet

Setting	Meaning
Follow-up after minutes	After this much silence, Teloring generates and sends a follow-up. 1–1,440 minutes.
Follow-up message sample	Guidance for that follow-up.
Give up after additional minutes	Grace period after the follow-up. 1–1,440 minutes.
No-response closing sample	Guidance for the closing sent before the action runs.
Action	Resolve, do nothing, return to the waiting line, or assign to a human.

If every mandatory field was already collected, reaching the follow-up time counts as a quiet success instead of abandonment.

The customer refused or got stuck

The AI can report that the customer will not provide the information, or that collection is going in circles. Choose the action for that outcome.

Escalation reasons

Six reasons can be configured: asked for agent, angry or frustrated, upsell opportunity, churn risk, forbidden topic, and low confidence. For each one, set whether it is enabled, an optional handoff message sample, and the routing action (with a target person if relevant).

An off switch does not stop detection

Turning a reason off disables its handoff handler, not the AI's ability to reach that outcome. A disabled reason can end the session with no routing at all. Keep the reasons you care about enabled, each with an explicit action.

Channels & Hours

The tab is named Channels & Hours, but today it only exposes working hours and after-hours behaviour. There is no per-inbox control here.

Setting	Meaning
Timezone	An IANA timezone such as Asia/Jerusalem or Europe/London. An invalid value falls back to UTC.
Day switch	Marks that weekday as an operating day.
Start and end	One operating interval for that day, inclusive of both ends.

Two behaviours worth knowing:

• If every day is switched off, the AI Agent is treated as available 24/7. It does not mean “closed all week”.
• Only one same-day interval is supported. An overnight range such as 22:00–06:00 is not treated as an overnight shift.

After-hours behaviour	Result
Do not respond	Skips the immediate AI reply outside working hours.
Respond normally	Runs the normal AI turn anyway.
Respond with after-hours message	Sends your fixed after-hours text instead of a generated answer.

warning

The scheduled follow-up job does not re-check working hours. A profile set to Do not respond can still send a follow-up later, outside hours. Do not rely on it as a guarantee of total silence.

What happens during an AI session

1. Someone assigns an open conversation to an active AI Agent.
2. Teloring checks the profile is active and snapshots it into a session.
3. The AI greets the customer when appropriate.
4. Incoming customer messages go to the AI Agent instead of AI Copilot.
5. For each turn Teloring searches the Knowledge Base, builds the instructions, and runs the model.
6. The AI answers, saves a configured field, or reports a terminal outcome.
7. Teloring sends the generated follow-up, closing, or handoff text the profile calls for.
8. The configured action resolves, unassigns, assigns a human, or does nothing.

AI messages carry an AI Agent badge. Model requests, retrieval details, responses, provider failures, and fallbacks are recorded for Teloring administrators to review in internal tooling.

Recommended setups

A support AI Agent

1. Write a narrow job title and bio.
2. Select only languages your content and team actually support.
3. Connect a tested Knowledge Base.
4. Start on KB + general language.
5. Add forbidden commitments and a few short behavioural rules.
6. Keep the mode on Give info (or Both if it must collect).
7. Send success to End and resolve.
8. Send failure and escalation to the waiting line or a monitored person.
9. Set working hours and test after-hours behaviour.
10. Run a full test conversation before going live.

A lead-collection AI Agent

1. Mode Both.
2. Add only the fields the business truly needs.
3. Use Customer, Contact, or Conversation targets.
4. Mark the critical ones Mandatory.
5. Turn Overwrite if present off for fields that must not replace verified data.
6. Route completed collection to a human, or resolve it — whichever matches your process.

---

Grouping agents into Teams

Both human agents and AI Agents can be grouped into Teams — the second page under Agents, Teams & Roles.

A team lets you assign a conversation to a group instead of one person, and it changes who may pull a waiting conversation with Get next:

	Effect on this page
Membership	Set on the Teams page, not in the agent form. An agent can be in many teams.
Human vs AI members	Both are allowed. A team's Only real agents rule excludes AI Agents and removes any already in it.
Auto-assignment	A team can hand a conversation to a random online member. AI Agents count as always online.
Deactivating an agent	They are silently dropped from teams the next time a team is saved.
Deleting an agent	They are removed from every team automatically.

Department is not a team. It is a free-text label on the agent profile, used for display and search only. Team membership is set on the Teams page.

Read the full guide: Teams.

---

What is not finished yet

These are real gaps, not things you are doing wrong. Plan around them.

Every item below was re-verified against the current build.

Item	What actually happens	Do this instead

| 2FA method: WhatsApp | Saved, but every code is emailed. | Leave it on Email. | | “2FA is not active yet” hint under the 2FA switch | Outdated text. Email 2FA works. | Ignore the hint. | | AI Get info goal mode | Never reaches the runtime's collection mode, so nothing is collected. | Use Both. | | AI custom object goal targets (object.<schema>.<field>) | Offered in the picker; the runtime does not write them. | Use Customer, Contact, or Conversation. | | Conversation priority / topic / notes targets | Write to conversation.custom_fields, not the real conversation controls. | Use them only as custom data. | | Goal data type file | Stores text. Nothing is uploaded or registered as a CRM file. | Collect a link, or use a form or attachment instead. | | Max consecutive AI messages | Goes into the prompt only. Unlike max turns and max duration, nothing enforces it. | Treat it as a hint. | | Escalation reason off switch | Disables the handoff handler, not the detection. The session can end with no routing. | Keep needed reasons on, each with an action. | | Escalation “other” fallback | Not in the editor, and hardcoded to no action. | Configure the six visible reasons and avoid no-action paths. | | Allowed inboxes for an AI Agent | The runtime honours the setting, but the editor has no control for it and clears it on every save, so it can never be used. | Control where an AI runs by only assigning it there. | | After-hours Do not respond | Stops immediate replies; scheduled follow-ups still go out. | Do not treat it as full silence. | | Add AI Agent name box | Still a plain browser prompt, unlike every other dialog on the page. | Nothing to do — it works. | | Active Agents statistic | Counts pending humans and active AI Agents too. | Read the Status column for real onboarding state. | | Pending humans in assignment pickers | They are technically active, so they can be picked before they finish registering. | Do not assign work to a Pending row. | | Deactivate and open sessions | Blocks new sign-ins only; an open session keeps working. | Also revoke sessions in Settings → Security & Login. | | Deleting a voice-enabled person | The voice extension is not deprovisioned. | Turn voice off first, then delete. | | Voice channel card | On/off only; the concurrent-call limit is backend-only. | Ask Teloring support to change the limit. | | Management buttons for non-admins | Some can still render; the server rejects the change. | Ask an admin to do it. | | AI Agent delete | There is no delete action for an AI Agent row. | Set the profile Status to Inactive. |

Team-specific limitations are listed in the Teams guide.

Troubleshooting

Someone is stuck on Pending

• Check they opened the newest email — invitations last 1 hour.
• Click Resend confirmation email to issue a fresh link (the old one stops working).
• Check spam filtering and that the address is spelled correctly.

Someone cannot sign in after a password reset

If an admin reset it, the old password stopped working right away — they must complete the newest link, which lasts 1 hour. If they reset it themselves, the old password still works until they finish. Either way, sending a new link cancels the previous one, so make sure they are using the latest email.

A new password is rejected

It must meet all five rules: 8+ characters, upper, lower, digit, and one special character. The set-password page shows which rule is failing.

The email-change link does not arrive

It goes to the address currently on file, not to the new one. Confirm that inbox is reachable. If it is not, an admin can create the correct profile and deactivate the old one.

An AI Agent does not appear in Assigned to

• Confirm its Status is Active.
• Reload the conversation after saving the profile.
• Confirm the conversation is still open.

An AI Agent ignores changes I just saved

The running conversation uses the snapshot from when its session started. Reassign it, or start a new conversation, to pick up the new profile.

The AI is not collecting a field

• Use Both, not Get info.
• Mark the field Mandatory if it must be asked.
• Check the target starts with Customer, Contact, or Conversation — not a custom object.
• Confirm the conversation is assigned to the AI and the session is still live.

The AI cannot answer from the Knowledge Base

• Confirm a Knowledge Base is selected.
• Confirm it has processed sources.
• Lower the confidence threshold for broader recall.
• Ask the same question on the Knowledge Base page to compare.
• In strict mode, missing context is meant to produce a refusal or a limited answer.

Voice cannot be enabled for someone

• Turn the account Voice channel on first.
• Confirm the profile is active.
• If provisioning reports an error, the voice server needs attention before retrying.

Related guides

• Teams — group agents and route conversations to a whole team.
• Conversations — assignment and the AI Agent queue.
• AI World — account-wide AI Copilot and AI feature switches.
• Knowledge Base — build and test grounded sources.
• Settings — account-wide 2FA, security, and session revocation.
• Voice inbox — voice channels and calling behaviour.