API 문서

Gemini API

Gemini 3.5–3.8 Flash and Gemini 4 Argon use /chat/completions. For Argon, use model ID google/gemini-4-argon. GET /models lists published models. For an unlisted model configured by your account administrator, use the exact model ID they provide with the same endpoint.

curl "https://tokenhub.store/api/v1/chat/completions" \
  -H "Authorization: Bearer th-your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemini-3.8-flash",
    "messages": [{"role": "user", "content": "Hello"}],
    "reasoning_effort": "low",
    "max_tokens": 1024,
    "stream": true
  }'

Supports streaming, system instructions, function calling, and base64 image data URLs. reasoning_effort accepts low, medium, or high; max_tokens includes thinking tokens.

For multi-turn tools, keep the complete assistant message, including tool_calls.extra_content or content_blocks, so thought signatures are preserved. Native Gemini endpoints are not exposed by TokenHub.

Gemini Flash and Argon output usage includes thinking tokens. Cached input is counted at the configured cache-read rate when available. See the model catalog for current prices; confirm rates for unlisted models with your account administrator.

Responses API 스트리밍 및 요금

POST /v1/responses의 response.completed는 응답 완료를 나타냅니다. 이 이벤트 수신 후 연결을 닫아도 성공한 요청이 취소로 변경되지 않습니다.

0을 포함하여 공급자가 보고한 토큰 수를 우선 사용합니다. 누락된 값만 추정합니다. 완료 전에 취소해도 처리된 토큰에 요금이 부과될 수 있습니다.

Account management API

Generate an access token in Settings → Account information. Send Authorization: Bearer YOUR_ACCESS_TOKEN and TokenHub-User: YOUR_USER_ID. New-Api-User is also accepted; if both user headers are supplied, they must match. Open settings

Access tokens manage your account. Continue using th- API keys for model requests.

export TOKENHUB_ACCESS_TOKEN="YOUR_ACCESS_TOKEN"
export TOKENHUB_USER_ID="YOUR_USER_ID"

curl "https://tokenhub.store/api/balance" \
  -H "Authorization: Bearer $TOKENHUB_ACCESS_TOKEN" \
  -H "TokenHub-User: $TOKENHUB_USER_ID"

curl "https://tokenhub.store/api/keys" \
  -H "Authorization: Bearer $TOKENHUB_ACCESS_TOKEN" \
  -H "TokenHub-User: $TOKENHUB_USER_ID" \
  -H "Content-Type: application/json" \
  -d '{"name":"My application","spendLimit":10}'

curl --get "https://tokenhub.store/api/v1/billing/usage" \
  -H "Authorization: Bearer $TOKENHUB_ACCESS_TOKEN" \
  -H "TokenHub-User: $TOKENHUB_USER_ID" \
  --data-urlencode "apiKeyId=YOUR_API_KEY_ID" \
  --data-urlencode "model=alibaba/qwen-turbo" \
  --data-urlencode "startDate=2026-09-07T00:00:00Z" \
  --data-urlencode "endDate=2026-09-08T00:00:00Z" \
  --data-urlencode "page=1" \
  --data-urlencode "limit=20"

Usage filters: apiKeyId is the ID returned by /api/keys, not the th- secret. model matches the full model ID exactly. Send startDate and endDate together as ISO timestamps with a timezone (inclusive boundaries). page defaults to 1; limit defaults to 20, maximum 100. Summary covers all matching records. Omit apiKeyId to query the account. An unknown key or a key owned by another account returns 404. API keys retain their existing account-wide billing access. /api/usage also accepts start_date/end_date, one-sided ranges, limit (default 100, maximum 1000), and offset (default 0).

Create: name=My application, spendLimit=10; omit spendLimit for the shared wallet. PATCH accepts name, isActive, and action (set_limit, add_budget, clear_limit, reset_spent); set_limit/add_budget also need a non-negative value. set_limit resets spent; add_budget preserves spent for an existing cap. GET /api/keys/:id returns the full model key. Deleting a key removes its UsageLog association; use isActive:false to retain usage filtering by key.

API 키를 이름이나 접두사로 검색하고 페이지당 20개씩 볼 수 있습니다. 새 키는 첫 페이지에 표시됩니다. 다른 창이나 API에서 만든 키를 보려면 목록을 새로고침하세요. API Keys

One active access token per account. It remains valid until regenerated or revoked; the database stores only its hash. Token lifecycle endpoints GET/POST/DELETE /api/user/access-token require a browser login. Tokens cannot manage their own lifecycle or authorize admin, password, account-deletion, or model-generation endpoints. Invalid tokens or mismatched user IDs return 401. Existing browser-session authentication continues to work.