Overflow & Exit Destinations

A queue can’t hold a caller forever, and sometimes shouldn’t admit them at all. Exit destinations define where a call goes instead of bridging to an agent. SIP.IO models five queue exits in two phases, plus a single ring-group exit, each a normal route target:

• At enqueue (before the caller ever waits): queue_full and ewt.
• While waiting (runtime outcomes): timeout, abandon, and no_agents.

In-queue exits

These fire while a caller is already holding:

{
  "id": "q_01j160r23gw5zkpfjme3fwzs5k", "name": "Support",
  "max_wait_sec": 600,
  "timeout_dest_kind": "voicemail", "timeout_dest_id": "vm_01jg8asp59ps9rtt7azhrhvcre",
  "abandon_dest_kind": "flow",      "abandon_dest_id": "flow_01jzqag8s7k8tan2475cshesh4",
  "no_agents_dest_kind": "flow",    "no_agents_dest_id": "flow_01jzqbs3hn9rb3sp06ypecjn0t"
}

Exit	Fires when	Typical destination
timeout	The caller waited longer than max_wait_sec.	Voicemail, a callback offer, or another queue.
abandon	The caller hung up while waiting.	A flow that triggers a callback or logs the abandon.
no_agents	There is no viable agent at all (none registered + logged in).	An after-hours flow, an overflow queue, or voicemail.

Each exit is a *_dest_kind / *_dest_id pair: point it at a voicemail, flow, queue, user, forward, or hangup.

Deflect at enqueue: queue_full and ewt

The best overflow is the one that never makes the caller wait. These two exits are evaluated the moment a caller would join; if they fire, the caller is deflected before holding at all.

{
  "max_queued": 25,
  "queue_full_dest_kind": "voicemail", "queue_full_dest_id": "vm_01jg8asp59ps9rtt7azhrhvcre",
  "ewt_threshold_sec": 180, "ewt_aht_sec": 240,
  "ewt_dest_kind": "forward", "ewt_dest_id": "fwd_01j3nnnyejaj858ffg9t764m43"
}

Exit	Fires when	Config
queue_full	Callers already waiting ≥ max_queued.	max_queued (NULL = unlimited).
ewt	The new caller’s estimated wait would exceed ewt_threshold_sec.	ewt_threshold_sec + ewt_aht_sec (handle-time estimate, default 180).

Estimated wait time is computed at the moment of joining:

ahead = max(0, waiting + 1 − ready_agents)
EWT   = ahead × ewt_aht_sec / viable_agents      (seconds)

Idle (ready) agents collapse ahead toward zero, so a queue with spare capacity reports an EWT near 0 and admits normally; a deep queue with few agents trips the ewt exit and deflects to, say, a callback or voicemail. Both checks fail open: a transient read error admits the caller rather than wrongly turning them away.

”No viable agent” vs. “all agents busy”

The distinction between timeout and no_agents is what makes the routing feel intelligent:

• All agents busy: members exist and are reachable, just occupied. The caller waits, and if nobody frees up in time, exits via timeout.
• No viable agent: no member is reachable and logged in. There’s no point waiting out the full max_wait_sec for help that can’t come, so the caller overflows immediately via no_agents.

Viability is “any member reachable and not offline”: a busy or paused agent still counts (they’ll be free soon), but an empty or fully-logged-out queue does not. See Queues & ACD.

caller would enqueue
   │  ── evaluated at join ──
   ├─ queue already at max_queued? ──▶ queue_full_dest   (deflect, no wait)
   ├─ estimated wait > threshold?  ──▶ ewt_dest          (deflect, no wait)
   │  ── now holding ──
   ├─ no viable agent at all?      ──▶ no_agents_dest     (immediately)
   ├─ caller hangs up while waiting ──▶ abandon_dest
   └─ waited past max_wait_sec     ──▶ timeout_dest

Ring-group exit

A ring group has a single exit_dest_* that fires when the ring times out with no answer.

Chaining exits automatically

When a number points directly at a queue or ring group, the platform synthesizes a flow and chains the configured exits into it, so “DID → queue, on timeout → voicemail” becomes a real multi-node graph with no flow-building required. Read how synthesis works.

Business hours are a separate gate

Exit destinations handle runtime outcomes (waited, abandoned, nobody home). Being closed is handled earlier by business hours attached to the queue: a closed/holiday schedule overrides the route into the queue before the caller is ever enqueued. Use business hours for “we’re closed”; use no_agents for “we’re open but nobody’s logged in right now.”

Recommended pattern

A robust support queue usually wires all three:

• no_agents → an after-hours/overflow flow or voicemail (nobody’s available),
• timeout → voicemail or a callback offer (waited too long),
• abandon → a flow that captures the number for a callback (they gave up),

plus a business-hours schedule on the queue for the closed/holiday case.