Clarifai
AIRun computer vision and AI predictions with Clarifai. Agents classify images, video, audio, and text against models and multi-step workflows, manage inputs stored in apps, search media by visual similarity or concept, and browse available models.
What This Integration Enables
Computer vision output is a probability, not a verdict, and most automation failures with it come from pretending otherwise. Wire a classifier's confidence score directly to an irreversible action and you will eventually publish what should have been blocked, or block what should have been published, at scale. Clarifai gives FlowRunner agents the full vision toolkit: public and custom models, multi-step workflows, an indexed input store, and search by visual similarity or concept. FlowRunner supplies the missing discipline: routing by confidence, so the model's certainty determines whether the next step is automatic or human. - Run predictions against public community models or your own, on images, video, audio, and text - Chain models with Clarifai workflows for detect-then-crop-then-classify pipelines - Build a searchable media library by adding labeled inputs to an app - Find media by visual similarity to a reference image or by concept, with ranked scores - Browse models and concepts to discover capabilities before wiring a prediction step The result is a division of labor that holds up in production: models watch everything, agents act on the confident cases, and the [human-in-the-loop](/concepts/human-in-the-loop) step catches the band where being wrong costs the most.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Moderation that scales without numbing the moderators
Every upload to the platform runs through Predict With Workflow. High-confidence passes publish immediately; high-confidence violations quarantine with the evidence attached. The middle band, and only the middle band, reaches a moderator through [Slack](/integrations/slack) with regions and scores rendered inline. Each ruling is labeled back onto the input with Add Input, so the review queue is also the pipeline that makes the next model version better at exactly the cases this one found hard.
The media library that indexes itself
When a new file lands in [Google Drive](/integrations/google-drive-api), the agent passes it to Predict With Model with general-image-recognition, logs the file and its recognized concepts to [Airtable](/integrations/airtable), and files the image into the Clarifai app with Add Input. Months later, a designer needs "something like this hero shot": Search Inputs takes the reference image and returns ranked visual matches from everything the company has ever produced, no tagging sprints required.
Inspection photos with a trained eye
Field technicians upload site photos after every job. A custom model in the company's app, trained on labeled examples of correct and defective installs, scores each photo through Predict With Model. Clean installs close the work order automatically; suspected defects open a follow-up with the photo and score attached, and an operations lead confirms before a crew is dispatched. List Concepts keeps the label taxonomy visible, so the categories the model recognizes stay aligned with the categories the business actually tracks.
Human-in-Loop Highlight
Delete Input cannot be undone, and in a Clarifai app an input is rarely just a file: it is often a labeled example that the custom model's accuracy was trained on, and a searchable record of a past decision. So the library-hygiene agent that prunes the app never deletes on its own. When it finds inputs that look like duplicates, test uploads, or mislabeled examples, it assembles the case: the input, its labels, its role in past training runs, and what Search Inputs still finds it matching. A person approves each purge, because deleting a mislabeled training example also deletes the evidence you would need to understand what the model learned from it. The sweep finds candidates at machine speed; the decision about what the model gets to forget belongs to whoever owns the model.
Agent Capabilities
10 actionsPredict
2- Predict With Model Runs a Clarifai model against a single image, text, audio, or video input and returns the predictions: concepts, regions, embeddings, or generated text depending on the model. Accepts a public URL, base64 string, or FlowRunner file. Works with public community models and your own.
- Predict With Workflow Runs a Clarifai workflow, a chain of models, against a single input and returns results from every node. The action behind multi-step pipelines such as detect, crop, then classify. Requires a User ID and App ID because workflows live inside an app.
Inputs
4- Add Input Adds an image, text, video, or audio input to a Clarifai app so it can be indexed, searched, and used for training, with optional concept labels for supervised learning.
- List Inputs Lists the inputs stored in an app with pagination. Used to browse or export what an app has accumulated.
- Get Input Retrieves a single input's details and processing status by ID.
- Delete Input Permanently deletes a single input from an app. Cannot be undone, which is why it sits behind this page's human gate.
Search
1- Search Inputs Searches an app's inputs by visual similarity to a reference image or by predicted or labeled concept name, returning ranked hits with similarity scores.
Models
2- List Models Lists Clarifai models with pagination and an optional name filter: the models in a configured app, or the public community catalog. The discovery step for Predict With Model.
- Get Model Retrieves a single model's full details by ID, including its type, description, and default version.
Concepts
1- List Concepts Lists the concepts, the labels and tags, defined in an app. The vocabulary a custom model recognizes and inputs can be labeled with.
Frequently Asked Questions
What can FlowRunner do with Clarifai?
FlowRunner agents can run Predict With Model, Predict With Workflow, and Add Input in Clarifai, plus 7 more actions.
Does connecting Clarifai to FlowRunner require OAuth?
No. Clarifai connects to FlowRunner with an API key, no OAuth flow required.
Can Clarifai trigger a FlowRunner workflow automatically?
Clarifai doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Clarifai
$100 in credits. No card required. Connect in minutes.