Pay only for what your agents actually use.
No seats. No tiers. No monthly minimum. Top up a wallet, and every chat turn and every container second is deducted in real time at roughly 2× the underlying vendor cost. Some accounts get free starter credit. Stop when you stop.
Fast & cheap models
Most agent traffic lives here. Snappy, low-cost, good enough for 80% of turns.
- bolt GPT-5 mini — $0.30 in / $1.20 out per 1M tokens
- bolt Gemini 2.5 Flash — $0.60 in / $5.00 out
- bolt Claude Haiku 4.5 — $2.00 in / $10.00 out
- bolt DeepSeek V3 — $0.55 in / $2.20 out
- info All prices include the 2× platform markup.
Frontier models
When the task actually needs a brain. Same wallet, same endpoint — just a different config field.
- psychology Gemini 2.5 Pro — $2.50 in / $20.00 out per 1M tokens
- psychology Claude Sonnet 4.6 — $6.00 in / $30.00 out
- psychology GPT-5 — $20.00 in / $60.00 out
- psychology Claude Opus 4.7 — $30.00 in / $150.00 out
- info All prices include the 2× platform markup.
Sandbox containers
Per-second compute for sandboxed agents — filesystem, shell, build pipeline included.
- memory shared-cpu-2x · 2 GB — $0.12/hr
- memory performance-2x · 4 GB — $0.42/hr
- memory performance-4x · 8 GB — $0.82/hr
- pause_circle Auto-pauses after 10 min idle. Resumes on the next chat.
- info Billed per second; only while the sandbox is running. 2× included.
How metered billing works
| Step | What happens | What it costs |
|---|---|---|
| 1. Top up | Add funds to your wallet via Stripe. Whole-dollar amounts, $5 minimum. | Face value. No fee on top-up. |
| 2. Chat | Each turn reserves the maximum it could cost, runs the model, then refunds the unused portion. | (input + output tokens) × model rate × 2 |
| 3. Sandbox | Sandboxed agents reserve their session cap on spin-up; actual seconds settle as the container runs. | seconds × machine rate × 2 |
| 4. Stop | No traffic, no debit. The wallet just sits there until your next call. | $0 |
| Empty wallet | The next chat or sandbox call returns 402 Payment Required before any vendor cost is incurred. |
Nothing — the call doesn't run. |
Frequently Asked Questions
Are there subscriptions or tiers?
add
No. There is one wallet per tenant, you top it up when you want, and every chat turn and container second is deducted in real time. Nimblesite is metered pay-as-you-go end to end — no monthly minimum, no seats, no plans, no overage tiers.
Where does the 2× markup go?
add
It pays for the parts of the agent loop you'd otherwise build: per-tenant memory in Postgres, multi-tenant isolation, the chat → tool-call → result loop, sandbox provisioning and lifecycle, the model abstraction layer, and ongoing tracking of new providers and pricing changes. The default markup is 2.0×; partner accounts may negotiate lower.
Can I get free credit to try it?
add
Yes — early accounts get a starter credit so you can build and test without spending. Email us with what you're building and we'll set you up.
What happens when my wallet hits zero?
add
The next chat or workspace call returns 402 Payment Required. No vendor cost is incurred — we never let a request run that the wallet can't fund. Top up and your existing conversations resume exactly where they left off; nothing is lost.
Where can I see what I've spent?
add
The dashboard shows your live wallet balance, lifetime top-up, and lifetime spend. Every reservation, settlement, refund, and top-up is in an append-only ledger you can paginate through via GET /api/v1/wallet/ledger. Nothing is hidden.
Do unused tokens or credits expire?
add
No. The wallet balance is yours until you spend it. There is no monthly reset.
Can I self-host?
add
No. Nimblesite is a proprietary hosted API — there is no public source, no Docker image, and no self-host path. Every integration calls api.nimblesite.dev with an API key. Enterprise customers who need data residency can discuss a dedicated single-tenant deployment; contact sales.
Try it on the starter credit.
Spin up an agent, run a few hundred turns, and see exactly what the bill looks like before you ever top up.