Start and experiment
1,000credit
60,000 VND
The same 60 VND/credit rate, with no package lock-in or long-term commitment.
Built for Claude Code and your existing SDK
Claude Code,billed in Vietnamese dong.
Top up from 10,000 d, get an API key, and run. Keep your Anthropic or OpenAI SDK — change only the endpoint and pay for what you use.
No monthly plan. No maintenance fee. Every request is recorded with its model, token usage, and actual cost.
30-second setup
NKReady
Claude Code → NapKey
Keep the workflow, change the gateway
# Keep Claude Code. Change only the gateway.export ANTHROPIC_BASE_URL="https://api.napkey.io.vn"export ANTHROPIC_AUTH_TOKEN="nk_live_..."claude20K d
Minimum top-up
2
API protocols
0 d
Maintenance fee
Why NapKey
We handle the hard part, you just call the API
Streams that survive long runs
The proxy layer holds SSE connections through multi-minute sessions. No re-buffering, no cutting mid-stream.
Pooled accounts with failover
Requests round-robin across multiple upstream accounts. When one fails, traffic moves on and you never see it.
One key, every model
Opus, Sonnet and Haiku share a single key. Switch models with one field in the body, not a new credential.
Integration
Change the endpoint in two lines
NapKey speaks both the Anthropic Messages API and OpenAI Chat Completions. Your existing client works as-is.
export ANTHROPIC_BASE_URL="https://api.napkey.io.vn"
export ANTHROPIC_AUTH_TOKEN="nk_live_..."
claudefrom anthropic import Anthropic
client = Anthropic(
base_url="https://api.napkey.io.vn",
api_key="nk_live_...",
)
message = client.messages.create(
model="claude-sonnet-4.6",
max_tokens=1024,
messages=[{"role": "user", "content": "Chao Claude"}],
)
print(message.content[0].text)import OpenAI from 'openai';
const client = new OpenAI({
baseURL: 'https://api.napkey.io.vn/v1',
apiKey: process.env.NAPKEY_API_KEY,
});
const res = await client.chat.completions.create({
model: 'claude-sonnet-4.6',
messages: [{ role: 'user', content: 'Chao Claude' }],
stream: true,
});
for await (const chunk of res) {
process.stdout.write(chunk.choices[0]?.delta?.content ?? '');
}curl https://api.napkey.io.vn/v1/messages \
-H "x-api-key: $NAPKEY_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{
"model": "claude-sonnet-4.6",
"max_tokens": 1024,
"messages": [{"role": "user", "content": "Chao Claude"}]
}'Read the full integration guide
Pricing
One credit rate across every model
No manual input, output, or cache-token conversion. NapKey records upstream-reported credits and deducts the wallet at a fixed 60 VND per credit.
Regular building
5,000credit
300,000 VND
The same 60 VND/credit rate, with no package lock-in or long-term commitment.
Higher volume
10,000credit
600,000 VND
The same 60 VND/credit rate, with no package lock-in or long-term commitment.
Published rate
1 credit = 60 VND
Credits come from actual usage reported by the upstream service. The dashboard records each request, model, reference token counts, credits, and settled VND for reconciliation.
NapKey wallet credits do not expire under the current policy. Settled usage always retains the rate effective when it was recorded.
Billing
Prepaid, metered to the token
Top up the wallet
Bank transfer, or scan a VietQR code with the amount and reference already filled in. Transfers reconcile automatically.
Call the API
Each request deducts the tokens it actually used at the model's rate. No rounding up, no per-request minimum.
Audit every line
Each request writes one usage row: model, input tokens, output tokens, cost. Reconcilable down to a single call.
Know this first
- Can I withdraw my balance?
- No. Balance can only be spent on API calls; it is not refundable to a bank account. Please weigh that before topping up.
- Does the balance expire?
- It does not. No maintenance fee and no inactivity fee.
- What is the minimum top-up?
- 10,000 d. You can increase the amount in 1,000 d steps.
- Do you issue tax invoices?
- Not at this stage. If you need a VAT invoice, please get in touch before topping up.
Developer trust
Evidence, not a trust badge
The useful question is not whether a gateway says it is secure. It is whether you can inspect health, correlate a request, audit its charge, and understand where data goes.
Operating principle
Every charge should trace back to one request. Every limitation should be written down before you top up.
NapKey exposes the mechanics developers can verify: a public liveness probe, request IDs, per-call usage records, explicit upstream boundaries, and visible key-revocation propagation state.
- Request correlation
- Billable calls return an x-request-id that matches the request ledger.
- Auditable charges
- Model, input, output, cache tokens, latency and cost remain visible per request.
- Controlled credentials
- Management APIs return masked identifiers. Disable and revoke actions expose their synchronization state instead of pretending propagation cannot fail.
- Bounded content retention
- Prompt payloads are excluded from request debug logs; Responses API storage is separately disclosed with its 30-day cleanup policy.
Compatibility contract
Keep your client. Keep your exit.
NapKey competes by reducing migration cost, not by inventing a proprietary SDK. Use established Anthropic and OpenAI request shapes, inspect the active catalog, and move away by restoring your endpoint.
Portable by design
Your integration should outlive your gateway vendor.
The Compatibility Center records what is accepted natively, what NapKey translates, and where behavior can still differ from an upstream provider.
Open Compatibility Center- 01Anthropic Messages
- Messages, streaming, tool use and token counting through familiar request shapes.
- 02OpenAI Chat
- Chat Completions requests translated onto the same NapKey data plane.
- 03Responses API
- Responses-style calls supported with explicitly documented storage boundaries.
Want to try it?
Put in 10,000 d, grab a key, change the endpoint. If it is not for you, that is all you spent.