1
Prototype
Create a staging key, call two candidate models and save outputs for review.
Capabilities
One gateway for Chinese model access, route control and usage governance
SmarToken is built for teams that want to evaluate and operate Chinese AI model routes without rebuilding their OpenAI-compatible client code for every provider.
2
Evaluate
Score quality, latency, token cost and failure modes on real workflow examples.
3
Constrain
Set budget, model scope and ownership for each service key before exposing it to users.
4
Monitor
Review usage, balance changes, 429s and route availability during the first production window.
Platform modules
What SmarToken helps teams operate
The capabilities below are most useful when they are attached to a concrete workflow, owner and evaluation set.
OpenAI-compatible model access
Use familiar Chat Completions semantics while testing Chinese model routes behind one public API surface.
- - One /v1 base URL for supported chat completion clients.
- - Public model catalog pages for DeepSeek, Qwen, Kimi, GLM, Doubao and other configured families.
- - Streaming and non-streaming request modes for chat UIs, agents and automation jobs.
Best verified by sending the same prompt set through two or three candidate model IDs.
Route switching and upstream management
Keep application code stable while admins manage upstream provider keys, route priority and fallback options.
- - Application teams switch the model field instead of rewriting SDK code.
- - Admin-managed routes can separate provider credentials from user-facing API keys.
- - Fallback planning can be handled at the route and application layer.
Best verified by testing a controlled fallback scenario before launch.
Budgeted API keys
Separate keys by service, environment or workload so unexpected usage is easier to contain.
- - Create different keys for staging, production, batch jobs and customer-facing features.
- - Review usage history by model, account and key where logs are available.
- - Rotate exposed keys without changing unrelated services.
Best verified by creating one low-budget staging key before production traffic.
Usage and wallet visibility
Connect model calls to wallet balance so teams can understand spend before scaling.
- - Input and output token pricing are tracked separately by model.
- - Wallet history helps explain balance changes and promotional credit usage.
- - Pricing pages and model pages give public reference points for planning.
Best verified by comparing expected token usage with logged usage after a test run.
Developer docs and migration guides
Give overseas teams English pages for model selection, OpenAI SDK migration and troubleshooting.
- - Quickstart samples cover cURL, Python and TypeScript style integrations.
- - Migration content explains how to move from OpenRouter-style routing.
- - FAQ and help pages cover API keys, billing, errors and production readiness.
Best verified by asking a developer to complete the first call without private help.
Campaign and activation workflows
Support promotional credits, activation codes and onboarding campaigns without changing the API contract.
- - Free-credit campaigns can introduce new users to real API usage.
- - Activation codes and balance adjustments can support operator-led onboarding.
- - Campaign rules should remain explicit about eligibility, deadline and withdrawal limits.
Best verified by checking wallet history after a campaign grant or redemption.
Readiness
Before you move real traffic
Model routing is only useful if the surrounding operational habits are clear. Use this checklist before the first production rollout.
- API keys are stored server-side and rotated when exposed.
- Each production workflow has an owner, model default and fallback route.
- The team has a small evaluation set for support, coding, education or document tasks.
- Cost expectations are checked against the pricing page and actual usage logs.
- Support tickets include timestamps, model IDs and request paths but never private keys.
Need a workflow-specific starting point?
Start with use cases if you are deciding what to build, tutorials if you are wiring the SDK, or pricing if the first question is budget.