Gemini AI
AIAdd Google's multimodal AI to your agent workflows. Agents generate content, analyze documents and images, extract structured data, and process audio and video files through Gemini's Files API.
What This Integration Enables
Gemini integration extends FlowRunner's AI capabilities with Google's multimodal model. Agents use Gemini for tasks that require understanding content beyond text: analyzing scanned documents, interpreting images, processing audio, and generating structured outputs from complex inputs. Combined with FlowRunner's other integrations, Gemini becomes the reasoning layer for document-heavy and multimedia workflows.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Scanned Invoice Extraction
A vendor submits a scanned invoice image rather than a digital PDF. Parseur cannot extract structured data from a low-quality scan. The agent uploads the image to Gemini using Upload File, then calls Generate Content with a structured extraction prompt: "Extract vendor name, invoice number, invoice date, line items, and total amount from this invoice. Return as JSON." Gemini returns the structured data. The agent validates it and creates the bill in the ERP. Scanned documents that used to require manual entry are handled automatically.
Contract Summary for Due Diligence
An M&A agent is processing a stack of contracts as part of due diligence. For each contract PDF, it uploads the document to Gemini and calls Generate Content: "Summarize this contract in 3 bullet points. Identify any unusual terms, termination clauses, or change-of-control provisions. Return as JSON with fields: summary, unusual_terms, risk_level." The agent stores the structured output in the Notion due diligence database. The deal team reviews AI-generated summaries with risk flags instead of reading every contract themselves.
Customer Feedback Classification
Customer support emails arrive in a shared inbox. The agent reads each email, passes the content to Gemini with a classification prompt: "Classify this support message as one of: billing_issue, technical_issue, feature_request, compliment, or other. Extract the key issue in one sentence. Return as JSON." Based on the classification, the agent routes the email to the appropriate team in Jira or Asana. Support triage happens before anyone opens the inbox.
Human-in-Loop Highlight
Gemini AI is a reasoning tool, not a decision-maker. When an agent uses Gemini to classify a contract as high-risk or to extract terms that might affect a business decision, the AI output is an input to the human, not a substitute for the human. The agent formats the Gemini analysis and routes it to the relevant stakeholder via Slack: "Gemini flagged a change-of-control clause in the [Vendor] contract. Here is the relevant section and the AI summary. Review before I proceed with the next step in the approval workflow." The human sees the AI's work. They decide what happens next.
Agent Capabilities
28 actionsFiles
4- Upload File Uploads a file to the Gemini Files API for use in content generation. Downloads the file from the provided URL, uploads it to Gemini, and polls until the file is processed and ready.
- List Files Lists files uploaded to the Gemini Files API with pagination support. Returns file metadata including name, size, MIME type, and processing state.
- Get File Info Retrieves metadata for a specific file uploaded to the Gemini Files API. Returns details including file name, MIME type, size, creation time, and processing state.
- Delete File Deletes a file previously uploaded to the Gemini Files API. The file will no longer be available for content generation after deletion.
Content Generation
3- Generate Content Generates content using a Gemini model with a text prompt and optional file reference. Supports configurable temperature, max output tokens, and response format (text or JSON).
- Generate Content (Advanced) Generates content with the full Gemini feature set: multimodal inputs (Gemini file references or media URLs sent inline), multi-turn history, Google Search grounding with citations, URL context, code execution, custom function declarations, structured JSON output via a response schema, thinking budget control for reasoning models, safety settings, context caching, and full sampling controls.
- Count Tokens Counts the number of tokens a prompt (and optional Gemini file references) would consume for a given model, without generating content.
Image Generation
1- Generate Image Generates or edits images with a native Gemini image model (the gemini-*-image / 'Nano Banana' family).
Speech Generation
1- Generate Speech Converts text to natural speech using a Gemini text-to-speech model, with a choice of 30 prebuilt voices and optional multi-speaker dialogue (assign a distinct voice to each named speaker in the transcript).
Video Generation
4- Generate Video Generates a video from a text prompt (and optional starting image) with a Veo model, waits for generation to complete, downloads the result, and saves it to FlowRunner file storage.
- Start Video Generation Starts an asynchronous Veo video generation job and immediately returns the long-running operation name, without waiting for completion.
- Get Video Operation Checks the status of an asynchronous Veo video generation operation. Returns whether the job is done, the video download URI(s) once available, and the raw operation payload.
- Save Generated Video Downloads a generated Veo video from its download URI (authenticating with your API key) and saves it to FlowRunner file storage, returning a public URL.
Embeddings
2- Embed Content Generates a numeric embedding vector for a text using a Gemini embedding model. Supports task-type optimization (e.g. retrieval, similarity, classification) and configurable output dimensionality for storage-efficient vectors.
- Batch Embed Contents Generates embedding vectors for multiple texts in a single request using a Gemini embedding model.
Models
2- List Models Lists all models available through the Gemini API with their capabilities, token limits, and supported generation methods.
- Get Model Retrieves detailed metadata for a specific Gemini model, including its description, token limits, supported generation methods, and default sampling parameters.
Context Caching
5- Create Cached Content Creates a cached content entry (explicit context caching) from text and/or uploaded Gemini files, tied to a specific model.
- List Cached Contents Lists cached content entries with their metadata (content itself is not returned). Supports pagination.
- Get Cached Content Retrieves metadata for a specific cached content entry, including its model, expiration time, and token usage.
- Update Cached Content Updates the expiration (TTL) of a cached content entry, extending or shortening its lifetime.
- Delete Cached Content Deletes a cached content entry immediately, stopping further storage charges. Generation requests referencing the deleted cache will fail.
Batch Processing
6- Create Batch Job Submits an asynchronous batch generation job at 50% of the standard API cost with a 24-hour turnaround target.
- Get Batch Job Retrieves the status and results of a batch generation job. When the job has succeeded, inline results are returned directly; file-based jobs return an output file name to download with 'Download Batch Results'.
- List Batch Jobs Lists batch generation jobs with their states and metadata. Supports pagination.
- Cancel Batch Job Requests cancellation of a pending or running batch generation job. Requests already processed before cancellation may still be billed.
- Delete Batch Job Deletes a batch generation job record. This removes the job from your list but does not cancel in-flight processing; cancel first if the job is still running.
- Download Batch Results Downloads and parses the JSONL results file of a completed file-based batch job, returning the parsed result objects.
Start building with Gemini AI
$100 in credits. No card required. Connect in minutes.