Usage Cascade (what to run when your Claude subscription is out)
The usage cascade is a short ordered ladder of other AI vendors that Omniscio falls back to when every Claude account in your pool has genuinely run out, or when the vendor a session is running on (GLM, DeepSeek, Kimi, MiniMax) hits its own usage limit or runs out of balance, instead of parking sessions in a waiting state for hours. You build it in Settings, it ships empty, and a session that left Claude comes home once capacity returns.
On this page
What it is
When every Claude account in your pool has hit its usage limit, Omniscio's normal behaviour is to stop: affected sessions park in a waiting state, a banner appears in the sessions sidebar, and everything resumes on its own when your subscription window resets. That is the right default — it never spends money you did not ask it to — but it means work stops, sometimes for hours.
The usage cascade lets you say, in advance and in your own order, what should happen instead. You build a short ladder. Each rung names one other AI vendor and how many sessions may use it at the same time:
| # | Fall back to | How many at once |
|---|---|---|
| 1 | GLM | No limit (default) |
| 2 | DeepSeek | At most 5 |
| — | wait for your subscription to reset (always last, not removable) | — |
When the subscription runs dry, each affected session walks that ladder from the top and takes the first rung that is switched on, set up (its credential is present), not already full, and not over its daily spend limit (Settings → Session → Spending limits). If no rung qualifies, the session waits exactly as it does today.
"Runs dry" means every one of your live Claude accounts is genuinely at its limit. If one account is merely busy and another still has room, Omniscio just moves the session to the account with room, exactly as it always has — the cascade does not fire, and nothing is billed.
It also catches a vendor running out
The same ladder is used when a session that is already running on one of these vendors runs out on it — GLM hits its usage limit, say, or your DeepSeek balance hits zero. Instead of waiting for that vendor to reset, the session walks your ladder from the top, skipping the vendor it is leaving (and any vendor it already ran out on), and carries on at the first rung that qualifies.
For example, with just DeepSeek on your ladder, a GLM session that hits GLM's usage limit keeps going on DeepSeek (Flash, DeepSeek's default model) with its conversation intact, and re-runs the step that was interrupted.
A few details:
- The vendor's own backups come first. A second GLM key and every other row of the model family's supply list (Settings → Accounts → Who pays & who serves — your own account, a reseller, Omniscio credits) are used before a session leaves its vendor.
- No bouncing. A session never walks back onto a vendor it ran out on until that vendor is back by the same sign the move back up uses, so it moves at most once per rung and then waits. The record lives on the session itself, so restarting the app does not forget it.
- A session goes back up when it can tell. A vendor that hit its own usage limit is tried again once the reset time it named has passed — or, if it named none, after 15 minutes (never sooner than 1 minute, never later than 6 hours). A vendor that ran out of balance takes its sessions back once the balance is funded again, and one that hit its daily spending limit takes them back the next day. See "Coming back up" below.
- Existing ladders pick this up — and an empty ladder still changes nothing.
It ships empty. Until you add a rung, nothing changes at all.
Where to find it
Settings → Accounts → Usage cascade. The card shows your ladder in order:
- Add a fallback… — a dropdown of the vendors you can use. Vendors that run through Omniscio's own Claude process are offered (DeepSeek, Kimi, GLM, MiniMax and the like), plus Codex, which has its own way back to Claude. A reseller such as DeepInfra is not offered as a rung: it is a row inside a model family's supply list, and an older rung that named one now runs on that model's family, whose list decides who serves it. Other engines with their own separate session manager — Gemini, Devin, Cursor and friends — are deliberately not offerable: a session placed on one could never come home.
- No limit / Limit how many at once — a small switch per rung, off by default. Leave it off and the fallback runs as many sessions as it needs. Switch it on and a number box appears, capping how many sessions may sit on that one fallback simultaneously.
- An on/off switch per rung — turn a fallback off without losing its place in your order.
- Up / down arrows — reorder. Order is the whole point: the first rung that works wins.
- A trash icon — remove the rung entirely.
- A "Bills per token" label on any vendor that charges per use, so a paid rung is never a surprise.
- A "Carries a shortened copy of the chat" label on a Codex rung — see below.
Above your fallbacks the card shows Claude as rank 1, marked Always first: it is not a rung you can move or remove, and your fallbacks are numbered from 2. Three short lines at the top say, in plain words, when a session moves, that its chat moves with it (Codex gets a shortened copy), and that it moves back up on its own.
The final rung — wait for your subscription to reset — is shown as a non-editable row so the full behaviour is visible rather than implied.
With Settings → Lab → Provider fallback on errors switched on, the same card gains three more controls:
- Also switch when a provider keeps failing — off by default. See "When a provider keeps failing" below.
- Also move Claude sessions when Anthropic keeps failing (you pay the vendor per use) — off by default, and only available once the switch above is on.
- Order to try fallbacks in — Your order (the default) or Cheapest first.
How it behaves
What happens to a session that falls back
- The session starts on the fallback vendor and keeps its conversation — the transcript and context carry over, because these vendors run through the same underlying CLI.
- It says so once in its own transcript, naming the vendor it moved to and why.
- On its next message after your Claude subscription has room again, it moves back to Claude automatically, restored to the exact model it was using before, and says so in its transcript ("Claude has capacity again — this session is back on Claude."). A session with a background command still running waits for the message after that, so the command is not cut off. Restarting a session also brings it home.
Coming back up
A session that moved down your ladder does not stay down. Whenever a provider ranked higher than the one it is on is available again, the session moves back up to the best one that is — even if that is not all the way home yet.
For example, with Codex, DeepSeek, Kimi on your ladder, a Claude session that ended up on Kimi moves to Codex as soon as a Codex account has room again, even while Claude is still out — and then to Claude once Claude has room.
Each provider is judged by the one signal that really says it is back:
| Provider | Back when |
|---|---|
| Claude | your subscription has room again (and, after a daily spending limit, the next day) |
| Codex | the reset time Codex named has passed and one of your Codex accounts has room |
| A vendor that ran out of balance | the balance is funded again |
| A vendor that hit its daily spending limit | the next day |
| A vendor that hit its own usage limit | the reset time it named has passed (15 minutes if it named none; between 1 minute and 6 hours) |
A few details:
- It happens at the session's next message, or when it is restarted or resumed. A move back to Claude also happens on any message, including automatic ones.
- The chat comes along, exactly as when it moved down; Codex gets a shortened copy.
- One line in the chat says so — "Claude has capacity again — this session is back on Claude.", or "DeepSeek is available again, so this session moved back up to DeepSeek."
- No ping-pong. If the provider it moved up to refuses straight away, the session moves down again the normal way, and that provider is not tried again until its own signal says it is back.
- The same sessions stay put as on the way down: coaching, Ask and review sessions, and cloud or SSH sessions are never moved onto Codex.
Falling back to Codex
Codex is a different program, so it cannot pick up Claude's own conversation file. Instead, Omniscio replays the chat into Codex as a shortened copy — the start and the most recent part of the conversation — and does the same in reverse when the session comes home. A few things differ from the other rungs:
- The Codex rung is skipped while every Codex account you have signed in is out of its own limit.
- Coaching and Ask sessions are never moved onto any fallback, and cloud or SSH sessions are never moved onto Codex (it runs only on this computer).
- A session that runs out on another fallback vendor is not moved down onto Codex; it can move UP onto Codex when Codex ranks above where it is and has room again (see "Coming back up").
- It comes home on its next message, or when it is restarted, once Claude has room again.
When Codex itself runs out
A session running on Codex — one you started there, or one the ladder moved there — keeps going when every Codex account it can use is out of its limit:
- Codex's own backup comes first. If you are allowed to use more than one Codex account and another one has room, the session switches to it, exactly as before.
- Then the ladder. If no other Codex account can take it, the session moves to the next rung of your ladder below Codex (DeepSeek, say), carrying a shortened copy of the chat, and re-runs the step that was interrupted. One line in the chat says it moved and where.
- Codex stops waiting. The session no longer sits waiting for Codex to reset, so it never jumps back onto Codex behind the ladder's back.
- Home is kept. A session that came from Claude still goes back to Claude once Claude has room. A session you started on Codex goes back to Codex once the reset time Codex named has passed and one of your Codex accounts has room again.
- It stays put when it must. Coaching, Ask, review, SSH and cloud sessions wait for Codex instead, and with nothing on your ladder below Codex the session waits exactly as it did before.
You do not have to do anything to bring sessions home, and nothing is left stranded on a paid vendor after your subscription recovers. A session you started on a vendor — GLM, say — goes back to it the same way once that vendor's own usage limit has reset.
When a daily spending limit is reached
If you set a daily spending limit — on an API-key Claude account, or on a pay-per-use vendor like DeepSeek or GLM — a session that reaches it no longer just stops:
- It moves to your next rung. When the turn that crossed the limit finishes, the session moves to the next rung on your ladder that is switched on, set up, and not over its own daily limit. One line in the chat says which provider reached its limit and where the session went.
- The chat comes with it. Your next message starts on the new provider and carries on the same conversation (Codex gets a shortened copy, as above).
- It still stops when there is nowhere to go. With an empty ladder, or every rung over its own limit or not set up, the session stops exactly as before.
- A limit you set on one session still stops it. The per-session spending limit is you limiting that one session, so it never moves the session — only a provider's daily limit does.
- It does not bounce back the same day. A session that left Claude because your API-key account reached its daily limit does not come home until the next day.
- It stays put when it must. Coaching, Ask and review sessions are stopped instead, as before.
When a provider keeps failing
Running out of quota is not the only way a provider lets you down: it can also return server errors, say it is overloaded, or drop the connection. Normally Omniscio retries the same provider a few times and then parks the session. With the Lab switch Provider fallback on errors on and "Also switch when a provider keeps failing" on, a session that has used up those retries moves down your own ladder instead, and the turn that failed is re-run there.
The rules are strict, because moving a session can cost money:
- Only failures that moving can fix. Server errors (5xx), overloaded answers and dropped connections. A timeout never moves a session (the first provider may still be working on it, and a second call could be charged twice). Neither does a computer that has no network, or a refusal that comes from Omniscio's own gateway being at capacity.
- Only after the retries are spent, never on a single failure.
- One move per turn, and never onto a fallback the session already tried during the same run of failures. A move does not give the turn extra retries.
- Only your own ladder. It never adds a vendor, switches a disabled one on or raises a cap.
- Claude sessions need their own switch. A session on Claude moves only if you also turn on "Also move Claude sessions when Anthropic keeps failing". It moves to the first fallback that is set up and switched on. Sessions that cannot be moved safely, such as cloud sessions and reviewer sessions, are left alone. Every such move puts a note in your inbox naming the vendor, with a button that opens Settings, Accounts, because that vendor bills you for every use.
- It says what happened. The moved session notes in its transcript that the first provider "kept failing" (never "ran out"). If every fallback has been tried or is not available, the session waits as before and one inbox card says a failing provider has no fallback left.
Order to try fallbacks in
By default the ladder is tried exactly as you ranked it. Cheapest first puts the switched-on fallbacks in ascending price for the model each would run, using the same prices as the cost screens; a fallback with no known price goes after the priced ones, in your order. This only reorders fallbacks you already switched on. It never adds one, enables one, changes a cap or changes a model. The order applies to every move down the ladder, whether the trigger was running out or a failing provider. There is no "fastest" mode, because Omniscio has no reliable speed data per provider.
Why there is no "cheaper Claude model" rung
An earlier design offered "drop to a cheaper Claude model" as the first rung. It was removed deliberately, because it cannot work at the moment it would be consulted.
Omniscio decides the pool is exhausted by looking at your overall 5-hour and 7-day usage windows — not at per-model limits. So by the time the cascade runs, a cheaper Claude model would be asking the same account that is already at its cap, and would simply fail again. Worse, Claude always counts as "set up", so the rung could never be skipped: it would be taken every time, announce itself, and deliver nothing.
If you want Omniscio to use a cheaper Claude model when usage gets tight, that is a separate, already-shipped feature that fires earlier — when you approach a threshold, rather than after you hit the wall. See the rate-limit model downgrade settings.
The promises this feature makes
- Nothing happens until you ask for it. An empty ladder behaves exactly like today.
- It only fires when a lane is genuinely out — every Claude account at its limit, the vendor a session runs on refusing it for its usage limit or an empty balance, a provider reaching the daily spending limit you set, or (only if you switch on "Also switch when a provider keeps failing") the provider failing repeatedly after its retries. One Claude account running out still just switches accounts, as it always has.
- A vendor's own backups come first, and a session never bounces between two vendors that are both out.
- A cap is opt-in and means "at most this many at once" — never "run fewer sessions". A rung starts with no limit; you switch a cap on if you want one. It never limits your Claude sessions.
- Money is never spent by surprise. A rung that bills per token is labelled where you configure it, and can only ever be reached because you put it there.
- Every session that left Claude comes home once your subscription has capacity again, and a session moves back up to any higher-ranked provider whose own signal says it is back.
- A full, broken or over-limit rung is skipped silently, never an error. A rung whose vendor has hit its daily spend limit is passed over, so a session is never moved somewhere it would be stopped after one turn. Nothing fails because a fallback was unavailable.
- You can always see it happened — a move, and the move back to Claude, are each announced in that session's transcript.
Turning it off
Two ways:
- Clear the ladder (or switch every rung off) in Settings. The feature becomes inert.
- Set
AMC_DISABLE_USAGE_CASCADE=1in the environment. This restores today's behaviour exactly while leaving your ladder saved, so you can switch it back on without reconfiguring.
There is also a second, inherited switch worth knowing about: the cascade lives inside the
capacity-park route, so AMC_DISABLE_CAPACITY_PARK=1 turns it off too, one layer up.
Related
- account-pool.md — how Omniscio picks between multiple Claude logins, and the exhaustion banner this feature takes over from.
- distribute-sessions-across-accounts.md — spreading sessions across the Claude accounts you already have, which happens before any of this.
Last verified 2026-11-16