Comprehensive Gemini integration supporting Veo 3 video generation, Gemini Flash text generation (Nano Banana), chat completions, and multimodal AI capabilities via the Google Gemini API.
Comprehensive Gemini integration supporting Veo 3 video generation, Gemini Flash text generation (Nano Banana), chat completions, and multimodal AI capabilities via the Google Gemini API. On Nagent, Gemini is exposed as a fully-configurable artificial intelligence integration that any agent can call — 8 actions, and no authentication authentication. No code is required to wire Gemini into your workflow — connect it once via the External Integrations panel and reuse it across every agent you build.
Agent builders use Gemini to automate the kinds of tasks artificial intelligence teams previously handled manually. Concrete examples — each one is a single agent step in Nagent — include:
Every action and trigger is paired with a structured input/output schema (visible in the sections below), so when you wire Gemini into Helix — our agentic agent builder — the editor knows exactly what each step expects and produces. Configure once, deploy anywhere across your Nagent agents.
Every operation an agent can call against Gemini, with input parameters and output schema. Drop these into any step of an agent built in Helix.
GEMINI_COUNT_TOKENSCounts the number of tokens in text using Gemini tokenization. Useful for estimating costs, checking input limits, and optimizing prompts before making API calls.
Input parameters
Text to count tokens for
Model to use for token counting. Must be a model that supports the countTokens method. Use the ListModels action to see available models and their supported methods.
Output
Data from the action execution
Error if any occurred during the execution of the action
Whether or not the action execution was successful or not
GEMINI_EMBED_CONTENTGenerates text embeddings using Gemini embedding models. Converts text into numerical vectors for semantic search, similarity comparison, clustering, and classification tasks.
Input parameters
The text content to generate embeddings for.
Embedding model to use. Options: 'text-embedding-004' (768 dimensions, default), 'gemini-embedding-001' (3072 dimensions, latest).
Optional title for the content. Use with task_type='RETRIEVAL_DOCUMENT' to improve embedding quality for document search.
Specifies the intended use case to optimize the embedding. Options: 'RETRIEVAL_QUERY' (search queries), 'RETRIEVAL_DOCUMENT' (documents to be searched), 'SEMANTIC_SIMILARITY' (text similarity), 'CLASSIFICATION' (categorization), 'CLUSTERING' (grouping), 'QUESTION_ANSWERING' (question-document matching). Note: Some task types like 'CODE_RETRIEVAL_QUERY' may only be supported by certain models.
Truncate the embedding to this number of dimensions. Only supported by 'gemini-embedding-001' model. Recommended values: 768, 1536, or 3072. Lower dimensions reduce storage but may affect quality.
Output
Data from the action execution
Error if any occurred during the execution of the action
Whether or not the action execution was successful or not
GEMINI_GENERATE_CONTENTGenerates text content or speech audio from prompts using Gemini models. Supports text generation models (Gemini Flash, Pro) and text-to-speech models with configurable parameters. Generated text is nested at results\[i\].response.data.text. Output may be wrapped in markdown fences (e.g., ```html...```) or preceded by explanatory prose; strip these before file writing or rendering.
Input parameters
Model to use for generation. Text generation models: 'gemini-2.5-flash' (default, fast & efficient), 'gemini-2.5-pro' (advanced reasoning), 'gemini-2.0-flash' (previous generation), 'gemini-2.0-flash-lite' (cost-optimized). Text-to-speech models: 'gemini-2.5-flash-preview-tts' (low latency), 'gemini-2.5-pro-preview-tts' (high quality). Note: TTS models require voice_name parameter and return audio data instead of text.
Top-k sampling parameter
Nucleus sampling parameter (0.0 to 1.0)
REQUIRED. The text prompt for content generation. This field must be provided. Example: 'Write a short poem about the ocean' or 'Explain quantum computing in simple terms'. For TTS models, include style instructions in the prompt (e.g., 'Say cheerfully: Hello!').
Available prebuilt voices for text-to-speech generation. Complete list of 30 official Gemini TTS voices as documented at: https://ai.google.dev/gemini-api/docs/speech-generation
Controls randomness (0.0 to 2.0)
Sequences where generation should stop
Safety filter settings
Maximum number of tokens to generate If response finishReason='MAX_TOKENS', output was truncated; narrow prompt scope or increase this value and regenerate.
System instruction to guide the model's behavior
Output
Data from the action execution
Error if any occurred during the execution of the action
Whether or not the action execution was successful or not
GEMINI_GENERATE_IMAGEGenerates images from text prompts using Gemini models (Nano Banana). Supports models: 'gemini-2.5-flash-image' (GA stable, fast), 'gemini-3-pro-image-preview' (Nano Banana Pro - advanced with 4K resolution, thinking mode, up to 14 reference images), and 'gemini-2.0-flash-exp-image-generation' (2.0 Flash experimental). Returns one image per call; images are uploaded to S3. Parse response at data.image.s3url or the text-type entry in data.content — prefer the URL to avoid base64 blobs. Always validate s3url before treating call as successful; a 200 response may contain only text with no image. Store s3url immediately as URLs can expire. Output formats are raster only (JPG/PNG/WebP); request PNG for transparency. Concurrent usage may trigger HTTP 429/RESOURCE_EXHAUSTED — keep concurrency ≤3 and use exponential backoff (1s→2s→4s, ~5 retries). NOTE NEVER EVER TRUE SYNC_TO_WORKBENCH IN RUBE_MULTI_EXECUTE_TOOL
Input parameters
Model to use for image generation. Options: 'gemini-2.5-flash-image' (GA stable, fast), 'gemini-3-pro-image-preview' (advanced with 4K, thinking mode), 'gemini-2.0-flash-exp-image-generation' (2.0 Flash experimental).
Top-k sampling parameter
Nucleus sampling parameter (0.0 to 1.0)
Text prompt for image generation Sensitive, trademarked, or explicit content triggers HTTP 400 (PROHIBITED_CONTENT or IMAGE_RECITATION) with no image returned — rephrase into neutral, policy-compliant language rather than retrying identical prompts.
Request timeout in seconds. Default is 300 seconds (5 minutes). Increase for complex prompts or high-resolution images. Minimum 120 seconds, maximum 600 seconds.
Output resolution (only for 'gemini-3-pro-image-preview'). Options: 1K, 2K, 4K.
Controls randomness (0.0 to 2.0)
Aspect ratio for generated image. Not supported by 'gemini-2.0-flash-exp-image-generation' model. Accepted values: '1:1', '4:5', '16:9', '9:16'. Unsupported strings will fail or silently default to 1:1.
Safety filter settings. List of objects specifying content categories to filter and threshold levels. Each setting requires 'category' (HARM_CATEGORY_HARASSMENT, HARM_CATEGORY_HATE_SPEECH, HARM_CATEGORY_SEXUALLY_EXPLICIT, or HARM_CATEGORY_DANGEROUS_CONTENT) and 'threshold' (BLOCK_NONE, BLOCK_LOW_AND_ABOVE, BLOCK_MEDIUM_AND_ABOVE, or BLOCK_ONLY_HIGH).
Maximum number of tokens to generate (max 32,768). For image generation, images consume tokens based on resolution: 1K/2K consume 1,120 tokens, 4K consumes 2,000 tokens. If set too low, the API may return MAX_TOKENS finish reason with no image. If not specified, the API uses its default which is sufficient for image generation.
System instruction to guide image generation behavior
Output
Data from the action execution
Error if any occurred during the execution of the action
Whether or not the action execution was successful or not
GEMINI_GENERATE_VIDEOSGenerates videos from text prompts using Google's Veo models. Returns an operation_name for tracking; pass it verbatim (no edits) to GEMINI_WAIT_FOR_VIDEO or GEMINI_GET_VIDEOS_OPERATION. Jobs take 30–180+ seconds; wait 10s before first poll, then poll every 10–30s (allow up to 12 min). Successful results include data.video_file.s3url — missing s3url means failure. If done=true but no video_file, check raiMediaFilteredReasons (safety block); revise prompt and regenerate. Text-only; cannot accept image inputs. Max ~3–5 concurrent jobs; 429 RESOURCE_EXHAUSTED requires exponential backoff. For retries, always start a fresh call — never reuse a failed operation_name.
Input parameters
Seed value for reproducibility. IMPORTANT: Only supported by Veo 3/3.1 models (VEO_3, VEO_3_FAST, VEO_3_1, VEO_3_1_FAST). VEO_2 does NOT support seed - using seed with VEO_2 will result in a validation error.
Veo model for video generation. Available enum values: VEO_3 (default, recommended), VEO_2, VEO_3_FAST, VEO_3_1 (newest), VEO_3_1_FAST (newest). Avoid preview model ID variants (e.g., '*generate-preview*') — they fail to produce downloadable URIs. Use only stable IDs: veo-2.0-generate-001 or veo-3.0-generate-001.
Text prompt for Veo video generation. Must be a non-empty string describing the video to generate.
Supported resolutions for video generation.
Supported aspect ratios for video generation.
Text describing content to avoid in the generated video (e.g., 'cartoon, drawing, low quality').
Supported video durations in seconds. Model-specific restrictions apply: - Veo 2: Supports 5, 6, 7, or 8 seconds (4 seconds NOT supported) - Veo 3/3.1 models: Supports 4, 6, or 8 seconds (5 and 7 seconds NOT supported)
Person generation safety settings for video generation. Model-specific restrictions apply: - Veo 2: Supports DONT_ALLOW, ALLOW_ADULT, and ALLOW_ALL - Veo 3/3.1 models: Only ALLOW_ALL is supported (requires allowlist access)
Output
Data from the action execution
Error if any occurred during the execution of the action
Whether or not the action execution was successful or not
GEMINI_GET_VIDEOS_OPERATIONDEPRECATED: Use WaitForVideo instead. Checks status of a Veo video generation operation. Use operation_name from GenerateVideos to track progress. Wait several seconds after starting GenerateVideos before first call to avoid OPERATION_NOT_FOUND. Poll at 10–30s intervals; use exponential backoff on HTTP 429 RESOURCE_EXHAUSTED; cap total polling at ~15 minutes. Complete when done=true AND a valid video URI is present; done=true without video_file indicates safety filtering blocked output — inspect raiMediaFilteredReasons and rephrase prompt. Video URL is at generatedSamples\[\].video.uri — persist promptly as URLs are time-limited. Keep concurrent polling to 3–5 parallel calls to avoid rate limits. If WaitForVideo times out, continue polling here using the same operation_name rather than starting a new GenerateVideos job.
Input parameters
The operation resource name from GEMINI_GENERATE_VIDEOS. Accepts either the full resource name 'models/{model}/operations/{operation_id}' or just the operation ID. If only operation ID is provided, it will be expanded to use the default model 'veo-3.0-generate-001'. Pass exactly as returned — do not truncate or edit. Never reuse an operation_name from a failed job; start a new GenerateVideos call instead.
Output
Data from the action execution
Error if any occurred during the execution of the action
Whether or not the action execution was successful or not
GEMINI_LIST_MODELSLists available Gemini and Veo models with their capabilities and limits. Useful for discovering supported models and their features before making generation requests. Before calling video generation tools, verify model availability here — preview Veo models (e.g., veo-3.0-generate-preview) may be unavailable or return missing video URIs; prefer stable models like veo-2.0-generate-001.
Input parameters
Maximum number of models to return per page (default 50, max 1000).
Token from a previous response's nextPageToken to retrieve the next page.
Filter models by name prefix (client-side). Leave empty to get all models.
Output
Data from the action execution
Error if any occurred during the execution of the action
Whether or not the action execution was successful or not
GEMINI_WAIT_FOR_VIDEOPolls a Veo video generation operation until completion, then downloads and returns the video as a FileDownloadable. Generation takes 30–120+ seconds (up to ~10–12 min); long waits are normal, not failures. On completion, the URL is nested at data.video_file.s3url — validate it is non-empty before downstream use. A done=true response without a valid s3url indicates safety filter rejection (check raiMediaFilteredReasons) or quota exhaustion — adjust the prompt and regenerate. On timeout, use GEMINI_GET_VIDEOS_OPERATION with incremental backoff before starting a new job. Keep parallel jobs to 3–5 to avoid 429 RESOURCE_EXHAUSTED errors.
Input parameters
The full operation name returned by GEMINI_GENERATE_VIDEOS. Format: 'models/<model-id>/operations/<operation-id>' where <operation-id> is an alphanumeric string (e.g., 'models/veo-3.0-generate-001/operations/m8dl4dtqqzg8'). IMPORTANT: Do NOT use placeholder values like '...' - use the exact operation_name string from the generate videos response. CRITICAL: Must be from a generate-video operation (VEO_2, VEO_3, VEO_3_FAST models), NOT generate-preview operations (VEO_3_1, VEO_3_1_FAST models). Do not reuse operation_name from a failed GEMINI_GENERATE_VIDEOS job — always start a new GEMINI_GENERATE_VIDEOS call for retried requests.
Output
Data from the action execution
Error if any occurred during the execution of the action
Whether or not the action execution was successful or not
No publicly available marketplace agent is found using this tool yet. There are 63 agents privately built on Nagent that already use Gemini.
Build on Nagent
Connect Gemini to any Nagent agent in minutes — no API key management, no boilerplate. Just configure and deploy.
The five questions agent builders ask before adopting a new integration.
Open the External Integrations panel inside Nagent (app.nagent.ai/externalIntegration), find Gemini, and click "Connect Now." You'll authenticate with no authentication (it’s public) — Nagent handles credential storage and refresh automatically. Once connected, Gemini is available to any agent in your workspace.
No. Nagent provides no-code integration for every tool. Once Gemini is connected, you configure its 8 actions directly in the agent builder UI — no API calls, no boilerplate, no schema management.
Helix — Nagent's agentic agent builder — lets you drop Gemini steps into any workflow visually. Pick an action (e.g., one of those listed above), fill in the inputs (Helix knows the required vs. optional schema for each parameter), and connect it to upstream/downstream steps. Triggers run as the entry point of an agent, so when a Gemini event fires, the agent kicks off automatically.
Every Gemini action and trigger ships with a fully-typed schema — input parameters with name, type, required flag, and description, plus the output payload shape. The schemas are documented in the sections above. Helix uses these schemas to validate your configuration at build time and to type-check the data flowing between steps.
Yes. While Gemini ships with 8 pre-built artificial intelligence actions, you can layer custom logic around them inside Helix — pre/post-processing steps, conditional branches, retries, or stitching Gemini together with other connected tools. For deeper customization, talk to our team about Nagent's Agentic AI Lab — forward-deployed engineers who build Gemini-based workflows tailored to your business.