Token Admin

Manage per-user API tokens (the sk-… keys used by the OpenAI-compatible /v1/... endpoints). UI: web/default-pro/src/views/token/Token.vue.

The page is the same for regular users and admins — admins see their own tokens (the API does not expose a cross-user listing). For cross-user inspection use /api/user/self (admin) or the log-search endpoint.

Data Model

model.Token (table tokens, key fields):

Field	Type	Notes
user_id	int	Owner
name	varchar(50)	Display name
key	varchar UNIQUE	Secret (clients prepend sk- when calling)
status	int	TokenStatusEnabled=1 / Disabled=2 / Expired=3 / Exhausted=4
expired_time	int64	unix seconds; -1 = never expires
remain_quota	int64	Remaining quota
unlimited_quota	bool	Bypass quota checks
models	text (comma separated)	Allowed models whitelist; empty = all
subnet	text	Client subnet whitelist (e.g. 10.0.0.0/8,192.168.0.0/16)
created_time / accessed_time	int64	unix seconds

Endpoints

Endpoint	Method	Auth	Description
/api/token/	GET	User	Current user's tokens (?p=, ?order=)
/api/token/search?keyword=	GET	User	Fuzzy match on name
/api/token/:id	GET	User	Single detail (must belong to caller)
/api/token/	POST	User	Create (key auto-generated by random.GenerateKey)
/api/token/	PUT	User	Edit (?status_only=true changes only status)
/api/token/:id	DELETE	User	Delete
/api/token/status	GET	relay auth	OpenAI-compatible credit_summary
/api/user/self/token	GET	User	Generate / regenerate the user's access_token (dashboard session token — distinct from token.key)

Implementation: controller/token.go.

Create Token — POST /api/token/

{
  "name": "my-bot",
  "expired_time": -1,
  "remain_quota": 500000,
  "unlimited_quota": false,
  "models": "gpt-4o,gpt-4o-mini",
  "subnet": "10.0.0.0/8"
}

• name ≤ 30 chars.
• subnet (when non-empty) is validated with network.IsValidSubnets (CIDR list).
• key is generated server-side; the response contains the full Token so the UI can show the key once.
• New tokens default to TokenStatusEnabled=1.

Enabling Preconditions

UpdateToken re-validates when flipping status to Enabled:

• If the previous expired_time <= now: returns "令牌已过期，无法启用，请先修改令牌过期时间，或者设置为永不过期".
• If remain_quota <= 0 and !unlimited_quota: returns "令牌可用额度已用尽...".

OpenAI-Compatible /api/token/status

GetTokenStatus is called by the relay and returns:

{
  "object": "credit_summary",
  "total_granted": <remain_quota>,
  "total_used": 0,
  "total_available": <remain_quota>,
  "expires_at": <expired_time * 1000>
}

expires_at is 0 when expired_time == -1. total_used is currently always 0 because tokens do not maintain their own used_quota counter.

Quota Deduction Path

The relay middleware reads the token on each /v1/chat/completions call:

• unlimited_quota=true → skip deduction.
• Non-empty models → only allow requests whose model is in the whitelist.
• Non-empty subnet → check the client IP matches.
• expired_time > 0 && expired_time <= now → treat as expired.
• remain_quota <= 0 (and not unlimited) → treat as exhausted.

Frontend Guide

• Top Base URL card shows /v1 with a copy button and a usage-guide modal.
• Add Token modal fields: name, expired_time (unix seconds or -1), remain_quota, unlimited_quota switch, models (multi-select / comma-separated), subnet (CIDR).
• Row actions:
  • View full key: modal that shows the plaintext (only available right after creation).
  • Edit: update all fields except key.
  • Enable / Disable: popconfirm.
  • Delete: confirmation.
• Clicking the eye icon next to models opens a modal listing the full whitelist.

Implementation Pointers

Concern	Location
CRUD handler	controller/token.go
AccessToken (user-level)	controller/user.go::GenerateAccessToken
OpenAI credit_summary	controller/token.go::GetTokenStatus
Middleware (auth / quota)	middleware/auth.go, relay/
Routes	router/api.go