Google Gemini - Atako

Let your agents use the Google Gemini API — content generation, embeddings, token counting and the model catalogue.

Connection

Authentication: API key.

Go to aistudio.google.com → sign in → “Get API key” → “Create API key”. Select a project (or create one), name the key, and copy it immediately. The key grants access to content generation, embeddings and model listing — no additional scopes required.See Google Gemini’s documentation.

Read actions (3)

Action	Description
`count_tokens`	Count the number of tokens in a text prompt using the model’s tokenizer. Arguments: model (string, e.g. “gemini-2.5-flash”); prompt (string). Returns: total_tokens (integer).
`get_model`	Retrieve metadata for a single model (token limits, supported methods, description). Arguments: model (string, e.g. “gemini-2.5-flash”).
`list_models`	List the models available to the API key. Arguments: page_size (integer, max 1000, default 50); page_token (string, for pagination).

Write actions (3)

Action	Description
`batch_embed_contents`	Compute embedding vectors for up to 100 texts in a single request. Arguments: model (string, e.g. “gemini-embedding-001”); texts (array of strings, max 100). Returns: embeddings (array of objects each with values: array of floats).
`embed_content`	Compute an embedding vector for a single text. Arguments: model (string, e.g. “gemini-embedding-001”); text (string); task_type (optional string — RETRIEVAL_QUERY, RETRIEVAL_DOCUMENT, SEMANTIC_SIMILARITY, CLASSIFICATION, CLUSTERING, QUESTION_ANSWERING, FACT_VERIFICATION); output_dimensionality (optional integer, reduced embedding dimension). Returns: embedding.values (array of floats).
`generate_content`	Generate a model reply for a text prompt (non-streaming). Arguments: model (string, e.g. “gemini-2.5-flash”); prompt (string, the user message); system_instruction (optional string, developer instructions); history (optional array of objects {role: “user”\|“model”, text: string} for multi-turn conversations); temperature (number 0–2); max_output_tokens (integer ≥ 1); response_mime_type (“text/plain” or “application/json”); response_schema (optional object, JSON Schema for structured output). Returns: candidates[0].content.parts[0].text and usageMetadata (prompt_token_count, candidates_token_count, total_token_count).

Permissions

Every action above must be explicitly granted to an agent before it can be used. See Permissions for the grant model and Security for how credentials are protected.