Troubleshooting

Most Studio problems fall into three buckets: the flow never started, the flow started but took the wrong path, or a block did not do what you expected. Work through them in that order.

Start here

1. Open Executions. This answers the biggest question first: did the flow run at all?
   • No runs at all → the trigger never matched. Continue with The flow never runs.
   • There are runs → open the newest one. The replay shows the path it took and the step panel shows exactly what each block was handed and returned. Continue with The flow runs but takes the wrong path.
2. Is the flow Live? Draft and Paused flows never run on real events.
3. Did you publish after your last edit? Editing changes the draft only.
4. Is the conversation owned by a human? Studio skips conversations assigned to an agent.
5. Rehearse it. For a chatbot, the simulator reproduces the conversation in seconds without sending anything. For everything else, Test Run walks the draft once with sample data.

Read the run instead of guessing

Most of the tables below can be answered directly from a run. Handed to this block in the step panel shows every setting with its variables filled in, so an empty variable or a wrong branch is visible rather than inferred. See Executions.

---

The flow never runs

Symptom	Likely cause	Fix
Nothing happens on a real message	The flow is still Draft, or the last edit was never published	Click Publish.
Nothing happens and the badge reads Paused	Someone paused the flow, or Teloring paused it because a dependency was deleted	Open ⚙ Flow Settings → Resume flow. If the flow card shows a warning badge, fix the block it names and publish instead.
Nothing happens, and the flow is live	The trigger's inbox filter does not include the inbox you tested from	Open the trigger and tick the right inbox — or untick everything to listen on all.
It worked, then stopped	An agent is assigned to the conversation	Studio stays out of conversations a human owns. Test with a fresh conversation.
It stopped for one specific customer	A previous run handed that conversation to a human	The handover sticks until the conversation is resolved or a flow runs End Session. Resolve it and try again.
The flow card shows a warning badge	A dependency was deleted — usually an analytics alert	Fix the block it names, then publish again.
A scheduled flow never fires	The schedule row's timezone, or the day/time, is not what you think	Check the row. Remember a Days interval on Recurring Schedule is capped at 24 hours.
A webhook returns an error	The request does not match the learned format	Turn Learn next request back on and send the new shape once.
A Conversation Changed / Object change flow does not fire when another flow or the AI Agent made the change	By default these triggers ignore changes made by automations, so a flow cannot loop on itself	Open the trigger and switch on Also react to changes made by automations. Chains still stop after 5 flows deep.
Only one of several matching flows runs	It usually does not — all matching live flows run. But once one hands the conversation to a human, the rest are skipped	Check whether an earlier flow has Assign AI / human agent on.

The flow runs but takes the wrong path

Symptom	Likely cause	Fix
A condition never matches	The Value / variable field has braces around it	Write message.content, not {{message.content}}, in a condition operand.
A condition never matches, part two	Using equals on free text	Switch to contains. Real people type "sales please", not "sales".
A numeric comparison always fails	The value is not a number	greater than and friends only work on numbers. A non-numeric value never matches.
Two branches ran when you expected one	Match mode is All matching branches	Switch to First match only, and order the rules from most specific to most general.
Nothing after the condition ran	No rule matched and there is no ELSE branch	Add an ELSE branch.
The flow is stuck after a question	A post-wait condition matched nothing and there is no ELSE	The flow waits for the next message and re-evaluates. Add an ELSE branch or an End Session block.
A branch runs but does nothing	Its output port is not connected	Drag a connector from that port.
A block never runs	It has no incoming connection	The block shows This block is not connected to anything. Wire it in.
Two messages arrived at the same time despite a Wait	Wait only delays its own branch	Siblings run immediately by design. Put the Wait on the shared path if you want everything delayed.

A variable is empty

tip

Open the run in Executions, click the step, and read Handed to this block. It shows the template you wrote next to the value it became, which usually identifies the problem immediately.

Symptom	Likely cause	Fix
{{some.variable}} shows up literally in a message	The key does not exist at that point in the flow	Drag it from Available Variables instead of typing it. If it is not listed, the block that produces it is not upstream.
{{contact.name}} is empty	The contact is brand new and has no name yet	Branch on contact.name with exists, or write a greeting that reads well without it.
{{conversation.id}} is empty	The trigger has no conversation — schedule, webhook, analytics alert, agent status	Supply a conversation ID yourself, from the payload or a lookup.
{{http.json.something}} is empty	Capture response is off, or the field is not in the response	Turn Capture response for next actions on, then send {{http.body}} to a Private Note to see exactly what came back.
{{answers.my_field}} is empty	The form field's Field ID is not my_field	Field IDs are the variable names. Check the form.
{{webhook.body.x}} is empty	The payload does not contain x, or the format was never learned	Use Learn next request and read the learned variable chips.
{{var.something}} is empty	Saved variables are per flow	A var.* value saved by one flow is not visible to another.

A block fails

Block	Symptom	Fix
Reply Message	Nothing is sent	Check the conversation ID resolves, and that the channel can still send (an expired WhatsApp window, a disconnected inbox).
HTTP Request	Always fails	Studio only calls public addresses. Localhost, private networks, and internal hostnames are blocked.
HTTP Request	Red JSON box	The body is not valid JSON after variables are filled in. Watch for a missing comma or an unquoted string.
Send Email	Choose which email inbox sends this email	Pick an inbox, or switch Send from to the Teloring system sender.
Send Email	No inboxes in the dropdown	Connect an email inbox in My Ring, or use the system sender.
Send Email	Recipients rejected	Maximum 25 addresses across To, CC, and BCC. Check for a typo.
Code	Fails every time	The code must return a value. It also has no network access — use HTTP Request for that.
Add / Update a record	Choose an object	Choose the object; the field list follows that choice. A flow cannot be published without it.
Add / Update a record	Validation failed	A value does not fit the field's type, a required field is empty, or a linked record does not exist. Check the object's fields; skipped_fields names what was not saved.
Delete a record	in_use	Other records still link to this one. Remove the links first.
Contact Update	A field is missing from the dropdown	Only the fields of the Contacts object (see the object builder) appear. Protected system fields cannot be written.
Agent Availability	Always Offline	With Specific agents selected and nobody ticked, the answer is always Offline — by design. Pick agents, or switch to Anyone in the account.
Agent Availability	An AI Agent is not counted	AI Agents never count as online; they hold no session.
Business Hours	Always Closed	Check the schedule's timezone and its weekly hours in Settings, and whether a holiday calendar is closing the day.

Voice problems

Symptom	Likely cause	Fix
Voice blocks are missing from the picker	The flow has no Incoming Call trigger	Only a flow that starts with Incoming Call is a voice flow.
Publishing is blocked	A messaging block is in a voice flow, or the reverse	Remove the block the warning badge names.
The prompt does not play	The file is not WAV	WAV only. 8 kHz mono PCM is the safe format.
Callers hear silence and the line drops	A path ends without a transfer or Hang Up	End every path with a closing prompt and Hang Up.
Pressing a key does nothing	That digit's port is not connected	Wire every digit, plus Timeout and Invalid.
Forward to Agent never connects	The agent is not signed in — there is no permanent phone registration	Pair it with Agent Availability, and always wire the Rejected output.
No recording was saved	Recording is off at the trigger	Recording is controlled only by Record this call on the Incoming Call trigger.
Test Run does not ring the phone	Test Run cannot place calls	Dial the number for real.

Building with AI (Hermes)

Symptom	Likely cause	Fix
Use AI to build is greyed out	The AI Studio switch is off for the account	An admin turns it on in AI World.
"This flow already has blocks. Create a new flow to build one with AI."	You opened the AI builder on a flow that already has a canvas	Hermes only builds into an empty flow, so it can never delete work somebody did by hand. Create a new flow.
Build my flow is refused, pointing at the editor	The canvas was edited by hand during the interview	Your edits win. Finish the flow in the editor, or start a new flow and a new interview.
The message box is disabled	The flow has been built — the conversation is closed by design	Edit the flow on the canvas, or create a new flow to run a new interview.
The conversation reopens instead of the canvas	The flow still shows ⚡ Continue with AI	That is the unfinished interview. Use the card's Edit button to reach the canvas instead.
Hermes picked the wrong inbox or team	It was told the wrong thing, or the resource did not exist yet	Open the block on the canvas and choose the right one. Create missing inboxes and teams before the interview.
A block Hermes made has a ⚠ badge	It needs something only you can supply — usually an uploaded WAV prompt or a WhatsApp template	Open the block and fill it in. The rail's Needs your attention list said so during the interview.
"Hermes didn't manage to answer that one. Send your message again."	The AI returned an answer Studio could not read. The turn was not saved, so nothing is out of step	Your message is put back in the box — press send again.
"That answer was too long for Hermes to finish."	A single reply exceeded what the AI can produce in one turn	Send the same thing in shorter pieces, or split it across two answers.
The interview stopped, saying it hit a limit	An interview is capped at 60 questions, and accounts have a short message throttle	Progress is saved. Wait a moment and continue, or finish the flow by hand in the editor.
I closed the tab and lost my place	You did not — every answer is saved	Reopen the flow from the Studio list and click the card.

The simulator

Symptom	Likely cause	Fix
💬 Test this flow is not on the panel	The selected block is not an Incoming Message trigger	The simulator only starts from that trigger. Click it on the canvas first.
"Only the Incoming Message trigger can be tested this way for now."	The block you started from is a different trigger	Use Test Run for those, or add an Incoming Message trigger.
The bot replies differently than it did in production	You are talking to the draft, not the published version	That is intended. Publish, then compare with a real run in Executions.
The simulator asks about a branch instead of choosing one	The branch depends on a value the simulation did not fetch — a webhook, a saved variable, a Code block, a CRM lookup	Pick what the outside system would have answered. See When the simulator asks you a question.
Business Hours or Agent Availability always asks	They read live system state, which a rehearsal must not consult	Answer with the case you want to rehearse. This is how you test after-hours behaviour at midday.
A branch button says nothing connected	That output port has nothing wired to it	Wire it. In production the flow would stop there.
Nothing appears in Executions after a simulation	A dry run is not a run	By design — the simulator records nothing. Use Test Run if you want a recorded run.
"This simulation has expired. Start a new one."	30 minutes of inactivity	Nothing was lost, because nothing was saved. Start again.
"The simulation stopped after a great many steps…"	The flow ran an unusual number of blocks in one turn	Look for a loop — two blocks pointing back at each other.
An email or webhook did not happen	Every outside action is skipped on purpose	The transcript names each one. To test real delivery, publish to a test inbox.

Things that surprise people

Behavior	Why it works that way
Editing a live flow changes nothing until you publish	Draft and live are separate copies on purpose.
Deleting asks you to type the flow name	Deletion is permanent and takes the versions and saved variables with it, so it needs a deliberate confirmation.
Pausing is not the same as deleting	A paused flow keeps every block and can be resumed in one click.
An empty filter means "everything"	An Incoming Message trigger with no inboxes ticked listens on all of them.
An empty agent selection means "nobody"	On Agent Availability, an unfinished Specific agents block must not behave like anyone.
A Wait does not delay other branches	Only the branch the Wait sits on is paused.
A saved variable is not shared between flows	var.* is scoped per flow so two flows cannot overwrite each other.
Duplicating a block does not copy its connections	You almost never want the copy wired the same way.
Sticky notes never run	They are editor-only comments and are not part of the published flow.
A simulation never appears in Executions	A rehearsal is not a run. Nothing is sent and nothing is recorded — that is the whole point of it.
A deleted block still shows in an old run	A run is replayed against a snapshot of the flow as it was, so the history stays readable after you rebuild.
Old runs disappear on their own	Each flow keeps the number of runs your plan allows, oldest out. A new run is never dropped to stay under the limit.
Building an AI flow ends the conversation	Once the blocks are yours to edit, a rebuild would have to overwrite your changes.

Getting more detail

Where	What it tells you
Executions	What actually happened on a real run: the path taken, every block's input and output, the branch chosen, the error, the timing. Start here.
The simulator	What your chatbot would say, reproduced in seconds, without sending anything.
Test Run	Whether the draft's shape is valid, without touching customers.
A temporary Private Note	Still useful for watching a value change across a long live flow, though the step panel in Executions now shows the same thing without editing the flow.
Version history (Version bar)	What changed, when, and who changed it — and the button to roll back.
Audit Log (Settings)	Who published, paused, resumed, restored, or deleted a flow.

Next

• Executions — the run history, and how to read one run.
• Chatbot simulator — rehearse a conversation without sending it.
• Testing & publishing — a safe release routine.
• Flow recipes — known-good flows to compare against.