Generate Gemini content

Generate content with the native Google Gemini API schema. The model in the URL is a Token360 public model ID; use GET /v1/models to discover currently available models.

Endpoints

POST /v1beta/models/{model}:generateContent — return one JSON response.
POST /v1beta/models/{model}:streamGenerateContent?alt=sse — stream Gemini JSON responses as Server-Sent Events.
POST /v1beta/models/{model}:countTokens — not supported in this release and returns 501 UNIMPLEMENTED.

The base URL for these endpoints is https://api.token360.ai. Do not append /v1 before /v1beta.

Authentication

Use either Google-compatible API key form:

Header: x-goog-api-key: sk-your-api-key
Query parameter: ?key=sk-your-api-key

The header form is recommended because URLs may be recorded by clients, proxies, and access logs.

Request body

The request follows the Gemini GenerateContentRequest wire format. Common fields include:

contentsContent[]

Required. Conversation content. Each item contains a role and one or more parts.

systemInstructionContent

Optional system instruction in Gemini Content format.

generationConfigobject

Optional generation controls such as maxOutputTokens, temperature, topP, topK, stopSequences, responseMimeType, and responseSchema. Support varies by model.

safetySettingsSafetySetting[]

Optional Gemini safety settings.

toolsTool[]

Optional tools, including functionDeclarations.

tools.typestring

Tool type, usually function.

tools.functionobject

Function definition.

toolConfigToolConfig

Optional function-calling configuration.

Token360 preserves Gemini field names and does not require a top-level model field because the model is part of the URL.

Non-streaming example

Shell

1curl https://api.token360.ai/v1beta/models/gemini-2.5-flash:generateContent \
2  -H 'x-goog-api-key: sk-your-api-key' \
3  -H 'Content-Type: application/json' \
4  -d '{
5    "contents": [{
6      "role": "user",
7      "parts": [{"text": "Explain vector databases in one sentence."}]
8    }]
9  }'

A successful response uses the native Gemini shape, including fields such as candidates, usageMetadata, and modelVersion when supplied by the upstream model.

Streaming example

Set alt=sse on streamGenerateContent:

Shell

1curl -N 'https://api.token360.ai/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse' \
2  -H 'x-goog-api-key: sk-your-api-key' \
3  -H 'Content-Type: application/json' \
4  -d '{
5    "contents": [{
6      "role": "user",
7      "parts": [{"text": "Write a two-line greeting."}]
8    }]
9  }'

Each SSE data: event contains a Gemini response object. Consume Gemini candidate parts in order and finish when the stream closes; do not expect the OpenAI [DONE] sentinel.

Multi-turn conversations

Gemini native requests are stateless. To continue a conversation, resend the earlier user and model contents with the new user turn:

JSON

1{
2  "contents": [
3    {"role": "user", "parts": [{"text": "My project is called Aurora."}]},
4    {"role": "model", "parts": [{"text": "Understood."}]},
5    {"role": "user", "parts": [{"text": "What is my project called?"}]}
6  ]
7}

Function calling

Declare functions under tools[].functionDeclarations. A model function call is returned as a functionCall part. Send the tool result in a later functionResponse part together with the preceding conversation contents. Both non-streaming and streaming endpoints preserve these Gemini-native parts.

Errors

Gemini endpoints return Google-style errors. Use the HTTP status as the primary failure signal:

JSON

1{
2  "error": {
3    "code": 400,
4    "message": "Invalid request",
5    "status": "INVALID_ARGUMENT"
6  }
7}

Routing, usage, and billing

Although the wire format is Gemini-native, Token360 still performs API-key authentication, SKU and provider routing, provider credential injection, error normalization, request archiving, usage statistics, and billing. Usage from generateContent and streamGenerateContent appears in the same Token360 usage and billing views as other LLM protocols.