Flows advanced configuration
The following table lists all advanced configuration parameters supported by Flows and references to relevant documentation sections.
| Parameter name | Description |
|---|---|
active_listening
|
Active listening mode |
continued_utterance
|
Continued utterances — a response is withheld only while the conversation stays on the same node |
error_message
|
Message to be played if LLM response couldn't be generated due to some error. Refer to Error message for detailed description of similar AI Agent functionality. |
empty_llm_response
|
Empty LLM responses |
gemini_api
|
Selecting the model API — applied per conversation node |
incomplete_turn
|
Incomplete turn detection — applied to conversation nodes |
init_conditions
|
Init conditions |
interrupted_response
|
Interrupted responses |
llm_add_period
|
Fixing punctuation |
llm_allowed_alphabets
|
Restricting the alphabets |
llm_logs
|
Detailed LLM logs for troubleshooting |
llm_replace_words
|
Replacing mispronounced words |
max_transitions
|
Maximum transitions between nodes |
max_turns_message
|
Max turns |
no_user_input_repeat
|
Handling user inactivity |
numbers_sequence
|
Numbers spoken in parts |
openai_api
|
Selecting the model API — applied per conversation node (flows use the stateless Responses API) |
post_call_analysis
|
Post call analysis |
prompt_log
|
Logging the agent prompt — the flow prompt is logged once: at the beginning of the conversation, or when the flow is handed the conversation |
reasoning_effort
|
Reasoning effort |
remove_symbols
|
Removing symbols |
replace_words
|
Replacing misheard words |
skip_after_play_finished
|
Skipping on once the response has played |
start_delay
|
Delaying the conversation start |
tool_certs
|
Custom tool certificates |
tool_logs
|
Detailed tool call logs |
tts_stream
|
TTS streaming |
user_gender_detection
|
Gender detection |
Maximum transitions between nodes
The max_transitions advanced configuration parameter configures maximum number of transitions between nodes. It may be used to limit total conversation length and prevents endless loops.
| Parameter | Type | Description |
|---|---|---|
max_transitions
|
int | Maximum number of transitions allowed in the flow. |
Example
{
"max_transitions": 30
}
Handling user inactivity
When user remains silent, NO-USER-INPUT events are generated, as described in Speech events. You can configure flow to repeat the last response upon one of these events via the no_user_input_repeat advanced configuration parameter.
Specify a list of numeric NO-USER-INPUT event occurrences that should trigger the repeat. For example, the following configures the flow to repeat last response upon the first inactivity event – NO-USER-INPUT-1:
{
"no_user_input_repeat": [ 1 ]
}
If you specify multiple indexes, e.g. [1, 2], response will be repeated upon multiple inactivity event occurrences (NO-USER-INPUT-1 and NO-USER-INPUT-2 in the above example).
| Parameter | Type | Description |
|---|---|---|
no_user_input_repeat
|
list[int] | NO-USER-INPUT event occurrences which trigger repeating the last LLM response. |
If you want to end or transfer the call upon user inactivity, create corresponding global node and use the following logical statement in global condition:
(user_utterance == "NO-USER-INPUT-2")
The above example uses second inactivity event (NO-USER-INPUT-2) as a trigger. Adjust the number based on your needs.
Delaying the conversation start
The start_delay advanced configuration parameter configures delay in milliseconds before the flow starts the conversation. It may be used to make sure the first response isn't played before the call is fully established.
| Parameter | Type | Description |
|---|---|---|
start_delay
|
int | Delay in milliseconds before the conversation starts. |
Example
{
"start_delay": 1500
}
This is the Flow counterpart of the AI Agent welcome_message_delay parameter.
Skipping on once the response has played
A conversation node with Skip response enabled proceeds to the next node as soon as its response is sent, while the caller is still hearing it. The next node may then call its tools or queue another message before the response has finished playing.
The skip_after_play_finished advanced configuration parameter makes such a node proceed to the next node only once its response has finished playing.
| Parameter | Type | Description |
|---|---|---|
skip_after_play_finished
|
bool | Proceed from a node with Skip response only once its response has finished playing. Default false. |
Example
{
"skip_after_play_finished": true
}
If the caller interrupts the response, the flow proceeds to the next node at once. If the end of the playback isn't reported within 30 seconds, the flow proceeds anyway, while the rest of the response is still playing.