Billing & credits

Credits

Conduix bills in credits. 1 credit ≈ $1 of model spend. You buy credits in advance via Stripe, and every successful API call deducts the per-token rate of the model that served the request. Failures don't cost anything.

Watch your balance at /dashboard/billing and your per-model burn at /dashboard/usage.

Topping up

Buy credits one-off via Stripe Checkout. Five tiers — $5, $10, $30, $50, and $100 — with progressive volume discount (per-credit rate drops at larger tiers). Each top-up creates a Stripe invoice you can download from /dashboard/billing.

Auto-reload

Skip the "account paused" emails. Configure auto-reload to charge your saved card whenever your balance drops below a threshold:

Auto-reload only runs after the threshold is crossed, never speculatively. Toggle on/off from billing settings.

Subscription plans

Four self-serve plans plus custom Enterprise. Full breakdown at /pricing:

Included credits roll into your wallet at the start of each billing period. Excess usage draws from your wallet (top-ups, auto-reload, or invoiced). Subscription credits reset monthly and don't roll over; top-up credits never expire.

Annual billing is available on Builder, Scale, and Pro at 17% off (two months free).

Enterprise — NET-30 invoicing

For volume customers who need PO numbers, custom MSAs, or DPA paperwork, Conduix offers invoice billing on NET-30 terms instead of card-on-file. Contact support@conduix.ai — we'll set up the customer record, mode-switch the org to billing_mode = "invoice", and route usage through monthly invoices.

Spend caps

Hard caps per organization stop runaway spend before it happens. Configured at /dashboard/settings:

When a cap is exceeded, requests fail with HTTP 402 spend_cap_exceeded — the provider call never runs. See Governance.

What an "insufficient credits" error looks like

{
  "error": {
    "message": "Insufficient credits. Current balance: 12 credits.",
    "type": "insufficient_quota",
    "code": "insufficient_credits"
  }
}

Top up at /dashboard/billing or set up auto-reload to avoid this.

Cache hits are free

Identical deterministic requests (temperature: 0, same messages, same model) served from cache are not billed. Look for x-conduix-cache: HIT in the response headers.

Credits & tokens FAQ

How many tokens do I get per credit?

It depends on the model. Premium models cost more per token; budget models cost less. Using a typical 50/50 input-output mix, approximate tokens per credit:

Why is the rate different per model?

We pass each model's token cost through directly. Claude Opus is the most expensive model on the market because it's the most capable; GPT-4o Mini is one of the cheapest because it's optimized for routine work. Conduix doesn't change those rates — we give you one API across all of them.

Input vs output tokens?

Output tokens cost more than input tokens — usually 3–5× more. This is standard across every LLM provider. A typical chat message (~200 input, ~200 output) is roughly balanced, but workload type matters:

Your dashboard at /dashboard/usage breaks down actual input:output ratio so you can see which workloads burn credits fastest.

What does a top-up actually buy me?

Concrete example using GPT-4o at ~400 tokens per chat message:

Route the same dollars to DeepSeek V3 (budget tier) and multiply everything by ~9× more usage.

Subscription vs top-up: what's the actual rate?

Subscriptions give a better per-credit rate in exchange for monthly commitment:

Rule of thumb: if you're consistently topping up, the next subscription tier is cheaper per credit. If your usage is bursty or seasonal, top-ups keep you flexible.

Do credits expire?

Subscription credits are an "included allotment" (think cell-phone minutes); top-ups are dollars-in-the-bank.

Can I see exactly how a call gets charged?

Yes. Every API call writes a usage log at /dashboard/usage with:

Retention by plan: Builder 7 days, Scale 30 days, Pro 180 days, Enterprise 7 years.

How do I estimate my monthly burn?

What happens when my credits run out?

API requests return 402 insufficient_credits. To resume:

Failed requests don't consume credits. If a model errors out or you hit a rate limit, you're not charged.