Switchboard actions
The catalog of changes Switchboard can plan and build, with their risk tiers, and what it deliberately won't do.
The catalog grows as Switchboard leaves preview. This page lists what's in it today.
Every change Switchboard can make is an entry in a code-owned catalog. Each entry defines the parameter schema, a risk tier, whether it can be undone, its cost, the plan it requires, a dry-run check, and the function that executes it. The planner, the validator and the executor all read the same catalog, so the model can't invent an action or relabel its risk.
#The catalog
| Action | What it does | Risk |
|---|---|---|
set_ring_order | Who rings on a number, in what order, for how long | Medium |
clear_ring_order | Remove a ring order | Low |
set_work_schedule | The workspace's business hours (for example 9:00–17:00 per day) | Medium |
route_number_to_ai | Inbound calls go to the AI agent | Medium |
route_number_to_team | Inbound calls ring the team | Medium |
route_number_to_voicemail | Inbound calls go straight to voicemail | Medium |
forward_number | Inbound calls forward to an outside number | Medium |
assign_number | Make a number a teammate's line | Medium |
set_voicemail_greeting | Voicemail greeting text and voice | Low |
configure_ai | The agent draft: greeting, voice (the 8 voices the app offers), jobs | Medium |
set_ai_enabled | Turn the agent on or off | Medium |
publish_ai | Push the agent configuration live | Medium |
create_call_flow | Draft an IVR menu, optionally based on an existing one ("add press 3 for billing") | Low |
publish_call_flow | Publish a call flow version | Medium |
attach_call_flow / detach_call_flow | Put a flow on a number, or take it off | Medium |
buy_number | Buy a local number (by area code) or a toll-free number. The admin picks the exact number | High |
upgrade_plan | Move to the plan a requested feature needs. Added by Ody only, never proposed by the model | High |
#Risk tiers
| Tier | Runs in preview? |
|---|---|
| Low | Yes |
| Medium | Yes |
| High | No, except upgrade_plan, which has its own confirmation card and a typed APPROVE. Other high-risk steps (buying a number) are shown on the plan so the full picture is visible; do them in Settings for now |
| Critical (anything not in the catalog) | Never |
The risk cap defaults to medium. Plans whose catalog cost is above the monthly spend cap ($20 by default) can't be approved; the plan upgrade is priced on its own card and doesn't count toward that cap. Approving a plan that still contains a high-risk step requires typing APPROVE.
#What it won't do
These are outside the catalog, and the model is told to answer them with an explanation and zero steps:
- Port numbers in or out
- Change the billing plan on its own (except the automatic
upgrade_planstep when a feature needs it) - Release a number
- Remove a teammate
- Change anything in another workspace
- Write the agent's knowledge base from scratch
- File 10DLC
Use the app, or the REST API where an endpoint exists.