Unofficial Google AI Studio API Documentation

Disclaimer: This documentation describes an unofficial, undocumented API used by the Google AI Studio web interface. It is not supported by Google and may change or break at any time without notice. Use of this API may be against Google's Terms of Service. Proceed at your own risk.

Authentication & Session Management

The API uses a combination of a primary authorization hash, an API key, and session cookies. These are temporary and must be obtained from an active AI Studio session. To maintain a session, you may also need to periodically generate a new access token.

1. How to Obtain Initial Credentials

  1. Open your web browser and navigate to https://aistudio.google.com/.
  2. Open the Developer Tools (usually F12 or right-click -> "Inspect").
  3. Go to the Network tab.
  4. In AI Studio, perform an action, such as sending a message.
  5. Look for a request to alkalimakersuite-pa.clients6.google.com in the network log. The most common one will be GenerateContent.
  6. Click the request and go to the Headers tab.
  7. Find and copy the following values from the Request Headers section:
    • Authorization: This is your primary authorization token (e.g., SAPISIDHASH ...).
    • X-Goog-Api-Key: This is your API key.
    • Cookie: This contains your session cookies.

It is highly recommended to store these as environment variables.

# Recommended: Set credentials as variables
export AUTH_TOKEN="SAPISIDHASH ..."
export API_KEY="AIzaSy..."
export COOKIE="AEC=...; NID=...; ..."

2. Maintaining the Session

The initial credentials may expire. The GenerateAccessToken endpoint appears to be used to refresh the session or obtain a new short-lived token. While its exact usage in subsequent calls is not fully clear from observation, it's likely a necessary step for long-running scripts.

Example curl Request:

curl 'https://alkalimakersuite-pa.clients6.google.com/$rpc/google.internal.alkali.applications.makersuite.v1.MakerSuiteService/GenerateAccessToken' \
  -X POST \
  -H "Authorization: $AUTH_TOKEN" \
  -H "X-Goog-Api-Key: $API_KEY" \
  -H "Cookie: $COOKIE" \
  -H 'Content-Type: application/json+protobuf' \
  --data-raw '["users/me"]'

Response:
The response is a JSON array containing a single base64 encoded string.

API Endpoints

The base URL for all API calls is: https://alkalimakersuite-pa.clients6.google.com/$rpc/google.internal.alkali.applications.makersuite.v1.MakerSuiteService/

List Models

Retrieves a list of all available models and their capabilities.

Example curl Request:

curl 'https://alkalimakersuite-pa.clients6.google.com/$rpc/google.internal.alkali.applications.makersuite.v1.MakerSuiteService/ListModels' \
  -X POST \
  -H "Authorization: $AUTH_TOKEN" \
  -H "X-Goog-Api-Key: $API_KEY" \
  -H "Cookie: $COOKIE" \
  -H 'Content-Type: application/json+protobuf' \
  --data-raw '[]'

Response:
The response is a single, large base64 encoded string. When decoded, it reveals a JSON array where each element is an array containing detailed information about a model (name, version, token limits, supported methods, pricing, etc.).

Example Decoded Response Snippet:

[
  [
    [
      "models/gemini-2.5-pro",
      null,
      "2.5",
      "Gemini 2.5 Pro",
      "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
      1048576,
      65536,
      [
        "generateContent",
        "countTokens",
        "createCachedContent",
        "batchGenerateContent"
      ],
      // ... many more properties
    ]
  ],
  // ... other models
]

Count Tokens

Counts the number of tokens in a given text prompt for a specific model.

Example curl Request:

curl 'https://alkalimakersuite-pa.clients6.google.com/$rpc/google.internal.alkali.applications.makersuite.v1.MakerSuiteService/CountTokens' \
  -X POST \
  -H "Authorization: $AUTH_TOKEN" \
  -H "X-Goog-Api-Key: $API_KEY" \
  -H "Cookie: $COOKIE" \
  -H 'Content-Type: application/json+protobuf' \
  --data-raw '["models/gemini-2.5-pro",[[[[null,"was steve jobs actually a psychopath?? or is that just a myth"]],"user"]]]'

Response:
The response is a base64 encoded string. When decoded, it reveals a nested array. The token count is the first integer value. For a decoded response of [17,[],[[[16],1]],null,null,[[1,17]]], the token count is 17.

GenerateContent API

This is the primary endpoint for all generative tasks, including text, chat, and multimodal interactions.

Response Streaming

All responses from GenerateContent are streamed in chunks.

Master Payload Structure

The GenerateContent request body is a complex JSON array with 5 main parts.

[
  "<model_name>",
  [ /* History & Prompt Array */ ],
  [ /* Tool Configuration Array */ ],
  [ /* Generation & Safety Config Array */ ],
  "<session_data_string>"
]

1. History & Prompt Array

This array contains the full conversation history.

Single Prompt:

[
  [
    [
      [
        null,
        "<your_prompt_here>"
      ]
    ],
    "user"
  ]
]

Multi-turn Chat:

[
  [
    [
      [null, "<first_user_prompt>"]
    ], "user"
  ],
  [
    [
      [null, "<first_model_response>"]
    ], "model"
  ],
  [
    [
      [null, "<second_user_prompt>"]
    ], "user"
  ]
]

System Prompt: A system prompt is set by making it the first turn in the history, with an empty model response.

[
  [
    [
      [null, "You are a helpful pirate assistant named Bob."]
    ], "user"
  ],
  [
    [
      [null, ""]
    ], "model"
  ],
  [
    [
      [null, "Ahoy! What be the weather today?"]
    ], "user"
  ]
]

Image Input (Multimodal): To include an image, you must first get an upload ID (process not documented here) and include it in the user turn alongside the text prompt.

[
  [
    [
      // The first element is the image data
      [null, null, null, null, null, ["<image_upload_id>"]],
      // The second element is the text prompt
      [null, "describe what you see in this image"]
    ],
    "user"
  ]
]

2. Tool Configuration Array

This array specifies which tools the model can use (e.g., Google Search, Code Interpreter). The numbers correspond to specific tools enabled in the AI Studio UI.

Example (enabling tools 7, 8, 9, 10):

[
  [null, null, 7, 5],
  [null, null, 8, 5],
  [null, null, 9, 5],
  [null, null, 10, 5]
]

3. Generation & Safety Config Array

This array controls the generation parameters. Based on observation, the structure is:
[null, null, null, <max_output_tokens>, <temperature>, <top_p>, <top_k>, "text/plain", ...other_nulls..., [<candidate_count>, -1]]

Example:

[null, null, null, 65536, 0.3, 0.95, 64, "text/plain", null, null, null, null, null, null, null, null, [1, -1]]

4. Session Data String

This is a long, opaque, base64-like string that appears to be a unique identifier for the request or session. It must be included.

Example: "!wcKlwprNAAZYy...G1jW20"

Full GenerateContent Example

This curl combines all the elements for a multimodal request.

# Note: The data payload is extremely long and sensitive to formatting.
# It is simplified here for readability.
DATA_PAYLOAD='["models/gemini-2.5-pro",[...history_and_image_array...],[...tool_config...],[...generation_config...],"<session_data_string>"]'

curl 'https://alkalimakersuite-pa.clients6.google.com/$rpc/google.internal.alkali.applications.makersuite.v1.MakerSuiteService/GenerateContent' \
  -X POST \
  -H "Authorization: $AUTH_TOKEN" \
  -H "X-Goog-Api-Key: $API_KEY" \
  -H "Cookie: $COOKIE" \
  -H 'Content-Type: application/json+protobuf' \
  --data-raw "$DATA_PAYLOAD"

This documentation was generated by analyzing network requests from the AI Studio web interface. It has not been exhaustively tested and is provided as-is. Contributions and corrections are welcome via pull request here.

Edit this page on GitHub