Ygg Lab Token — API Quota Gateway
Every model call, clear and in control — nine upstream provider types behind a single sk- key, with quota, logs and billing you own.
In short
Ygg Lab Token is YGG's self-built, OpenAI-compatible API quota gateway. It collapses nine upstream provider types into one sk- key and one /v1/chat/completions endpoint. Admins issue per-person keys with spending caps and expiry dates, every call's tokens and cost are logged individually — so teams stop juggling separate top-ups across vendors and always know who spent what.
What the agent takes off people's plates
Core capabilities
OpenAI-compatible unified gateway
POST https://token.ygg.com.cn/v1/chat/completions with Authorization: Bearer sk-… — point an existing OpenAI SDK at a new base_url and you are done. Nine upstream types supported: OpenAI, OpenAI-compatible, Anthropic, Azure, Gemini, DeepSeek, Volcengine, Tongyi Qianwen, AWS.
9 upstream types · 1 endpoint
Token and quota control
Members create and name their own keys (e.g. "paper experiment"), each with its own spending cap and expiry, switchable on or off at any time. Four visible states — active, expired, exhausted, disabled — so a runaway experiment burns only that one key's quota, never the whole team's wallet.
Per-call billing log
Every call records input tokens, output tokens, cost, model and timestamp. The dashboard rolls it up into today's requests, today's spend, cumulative usage, remaining balance, usage trend, model mix, recent calls and active keys — month-end reconciliation stops being an act of memory.
Multi-tenancy and group discounts
User management assigns quota, groups members, toggles open registration and sets the default balance for new accounts. Group discounts apply as a multiplier (1 = list price, 0.8 = 20% off), so one set of upstream channels can serve faculty, students and partners at once, each billed on its own terms.
Channel management and upstream failover
A single channel accepts a batch of pasted keys, a priority level, a self-hosted or proxy endpoint, automatic model-list discovery per key, and a one-click connectivity test. When an upstream raises prices or throttles you, reorder channel priority — the sk- keys and application code stay untouched.
Pricing, wallet and system integration
Model pricing is set per $/1M tokens with input and output priced separately, matching upstream conventions. The wallet supports package top-ups, redemption codes and invoice history, with WeChat Pay Native (public-key verification) and Alipay Face-to-Face RSA2. Downstream products can call POST /api/system/register with a system integration key to auto-provision an account by external user ID and mint a billable sk- token.
Side by side
Self-hosted token gateway vs each vendor's own console vs a public API reseller
| Dimension | Self-hosted Ygg Lab Token | Each vendor console, managed separately | Public API reseller |
|---|---|---|---|
| Cost roll-up | One wallet, split by person, key and model | N separate invoices and top-ups, merged by hand | One invoice, but rates and discounts set by the reseller |
| Quota control | Per-key cap and expiry, hard stop when exhausted | Usually account-level only — one runaway job drains the team | Sub-accounts exist, but granularity follows the reseller's product |
| Call-level audit trail | Every call logs input/output tokens, model, cost, timestamp | Log formats and retention differ per vendor, hard to compare | Logs exist but live with the reseller; export is limited |
| Switching upstream | Reorder channel priority; sk- keys and code untouched | Edit base_url and keys in each project, with regression cost | Depends on whether the reseller carries that model |
| Data ownership | Gateway and logs run on your own servers, fully self-hosted | Requests go straight to vendors; logs stay on their side | Both requests and logs pass through a third party |
Quantified before / after
Measured impact
Subjects and capability matrix
Covered modules
Dashboard
Spend · trend · model mix
API keys
Caps · expiry
Call log
Per-call tokens and cost
Wallet
Packages · codes · invoices
Channels
Multi-key · priority · health
Users
Quota assignment · groups
Model pricing
$/1M tokens · in/out separate
Group discounts
Multiplier · 1 = list price
How the agent and human supervision collaborate
System architecture (conceptual)
source
Lab member / downstream app
agent
Token gateway (sk- auth · metering)
human
Admin (channels / pricing / quota)
output
Nine upstream provider types
output
Call log / billing
Flow
- Lab member / downstream app→Token gateway (sk- auth · metering)Bearer sk- request
- Admin (channels / pricing / quota)→Token gateway (sk- auth · metering)Configure channels & quota
- Token gateway (sk- auth · metering)→Nine upstream provider typesRoute by priority
- Token gateway (sk- auth · metering)→Call log / billingMeter every call
- Call log / billing→Admin (channels / pricing / quota)Usage & cost review
Verifiable facts
Numbers and sources
Nine upstream channel types are supported: OpenAI, OpenAI-compatible, Anthropic, Azure, Gemini, DeepSeek, Volcengine, Tongyi Qianwen and AWS.
A single OpenAI-compatible endpoint is exposed: POST /v1/chat/completions, authenticated with Authorization: Bearer sk-….
Model pricing is denominated in $ per 1M tokens, with input and output priced separately.
Group discounts apply as a multiplier: 1 = list price, 0.8 = 20% off.
Default safety thresholds: a balance below $10 triggers a reminder; SMS codes are rate-limited to once per 60 seconds and five per 24 hours; passwords must be at least 8 characters.
Top-ups run on WeChat Pay Native QR (public-key verification mode) and Alipay Face-to-Face (RSA2 signing); SMS goes through Tencent Cloud SMS.
Delivery and pricing
Pricing
Included
- · Full self-hosted deployment — gateway core, member workspace and admin console — onto your own servers
- · Configuration for all nine upstream channel types (OpenAI, OpenAI-compatible, Anthropic, Azure, Gemini, DeepSeek, Volcengine, Tongyi Qianwen, AWS)
- · The complete admin console: channels, model pricing, group discounts, redemption codes and system integration keys
- · The user and quota system: sign-up and login, user groups, quota allocation, key management, call logs, dashboard and wallet
- · Deployment, admin training and one year of technical support
Provided by the client
- · Servers, cloud instances and database resources
- · Domain names and HTTPS certificates
- · API keys and account balances with each upstream vendor
- · WeChat Pay and Alipay merchant accounts, including their fees
- · A Tencent Cloud SMS account, signature/template filing and per-message fees
Add-ons
- · Technical support and version upgrades from year two (annual renewal)
- · Custom upstream adapters, SSO integration and deeper product integration (billed per person-day)
Scan for the full solution document
Scan with WeChat to reach the solution lead and request the complete documentation and deployment checklist.
Where it fits
When this approach does not apply
- The gateway aggregates and meters; it does not pay for inference. You still bring and top up your own API keys with each upstream vendor.
- Billing at $/1M tokens depends entirely on the admin entering correct input/output rates and group multipliers. Misconfigure them and the ledger is wrong — upstream price changes must be synced manually.
- It is an internal quota hub for a lab or team, not a public API marketplace. Selling access to the general public would require separate work on compliance, invoicing and fraud control.
- Every call takes an extra hop: client → gateway → upstream. Latency depends on both your server and upstream availability, so ultra-low-latency workloads need real measurement first.
Further reading