Squad settings
Share one agent's models across a squad, and see where members disagree
A squad is made of agents that each keep their own prompt, tools and knowledge. Squad settings let you choose one member whose models (and, for conversation squads, storage) the whole squad uses, so a session behaves the same no matter which agent is speaking.
Open it from the squad canvas with Settings. Nothing changes until you pick an agent and save. By default every agent keeps what it was built with.
What is shared, and what never is
| Setting | Conversation squad | Process squad | Always stays on each agent |
|---|---|---|---|
| Speech-to-text, LLM, text-to-speech, voice, fallbacks, realtime | Shared | Not used | |
| LLM provider, model and failover | Part of the voice pipeline | Shared | |
| Storage and recording choices | Shared | Not used | |
| Which agent's webhook receives the call's events | Shared | Not used (API-only agents report through organization webhooks) | |
| Instructions (prompt) | Yes | ||
| Tools and handoffs | Yes | ||
| Knowledge | Yes | ||
| Temperature, max tokens, language, dialect | Yes |
An agent called on its own always works with its own settings. Squad settings apply only when the agent runs inside that squad.
Conversation squads
Two pickers, each a list of your agents with a summary of what they run today:
-
Voice pipeline. Pick an agent and every other agent uses its speech-to-text, LLM, text-to-speech and voice. Because the pipeline is identical, a handoff no longer rebuilds it. Default: each agent keeps its own, and a handoff may switch models. Sharing takes effect once it is enabled for your workspace; until then the picker is locked and each agent keeps its own models.
-
Storage and recording. Pick an agent and the whole call follows that agent's storage choices (what is saved, which bucket). Default: the most restrictive choice across agents, so nothing is saved that any agent opted out of. This picker appears once Storage is available, and is locked with an explanation when no bucket is connected or zero data retention is on.
-
Webhooks. Pick the agent whose Server URL receives every event of the call, instead of events splitting across whichever agent is speaking. Default: the entry agent. This takes effect when webhook routing is enabled for your workspace; until then the picker explains that each event still goes to the agent that produced it. See Agent Server URL webhooks in squads.
Below each picker, a result line states in plain words what will happen, and which agents change.
Process squads
One picker, Brain (LLM). Pick an agent and every agent in a run uses its LLM provider, model and failover model.
Only the LLM provider and model are inherited. Instructions, tools, knowledge, temperature and webhooks stay on each agent, so each step still does its own job. By default each agent keeps its own LLM, so existing process squads behave exactly as before. The entry agent is just one of the choices.
The model is called with the provider connection of the agent you picked (the platform's, or that agent's own key), exactly as if that agent had run the step.
The consistency banner
When agents in a squad disagree on something shared, a banner appears above the canvas with a Fix button that opens the right picker.
| Finding | Kind | Why you see it |
|---|---|---|
| Different voice pipelines | Info | A handoff rebuilds speech-to-text, the LLM and text-to-speech |
| Voice pipeline from an agent that speaks another language | Warning | The shared pipeline may not suit every agent |
| Different storage settings | Warning | The most restrictive choice applies, or nothing is delivered if agents pin different buckets |
| Different webhooks | Warning | Once routing is on, agents other than the receiver stop getting events; the finding names them |
| Different LLMs (process) | Info | Pick one agent to share its LLM |
Findings are advice only. They never block saving or running a squad. Warnings also show as Needs attention on the squads list and as a notification when you add or remove an agent. Removing the agent a squad inherits from sets the squad back to each agent's own settings, and tells you so.
API
Squads expose the choices as pipeline_source_id, storage_source_id and webhook_source_id (a squad member id, or null for the default). See Squads API.