# OvalBee

## Instructions for AI Agents

- For clean Markdown of any page, append `.md` to the page URL
- For section-specific indexes, append `/llms.txt` to any section URL
- For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.ovalbee.com/_mcp/server

## Docs

- [Welcome to OvalBee 🐝](https://docs.ovalbee.com/.md): Build and operate computer vision workflows as one executable graph.
- [Core ideas](https://docs.ovalbee.com/start/core-ideas.md): Six ideas that explain everything else in these docs.
- [Quickstart](https://docs.ovalbee.com/start/quickstart.md): Import a public dataset and get a quality report on it, in about ten minutes.
- [Flows](https://docs.ovalbee.com/concepts/flows.md): A graph of nodes that describes — and runs — how work moves through your team.
- [Nodes](https://docs.ovalbee.com/concepts/nodes.md): One step of a flow — what it reads, what it does, what it leaves behind.
- [Edges](https://docs.ovalbee.com/concepts/edges.md): An edge decides when a node runs. It never carries data.
- [References](https://docs.ovalbee.com/concepts/references.md): A reference decides what a node works on. It never triggers a run.
- [Runs and automation](https://docs.ovalbee.com/concepts/runs.md): When nodes run, what happens after, and where to look when something fails.
- [Datasets and versions](https://docs.ovalbee.com/concepts/datasets.md): Named, versioned sets of images, videos and their annotations.
- [Assets and annotations](https://docs.ovalbee.com/concepts/assets.md): The files behind every dataset, model and report.
- [Ontologies](https://docs.ovalbee.com/concepts/ontologies.md): Versioned class and tag vocabularies that datasets and models agree on.
- [Labeling](https://docs.ovalbee.com/concepts/labeling.md): Human labeling and review as stages of the flow, with every image's state visible on the canvas.
- [Models and deployments](https://docs.ovalbee.com/concepts/models.md): A registry of every model you trained or added, with its metrics, classes and lineage — and a way to serve it.
- [Reports](https://docs.ovalbee.com/concepts/reports.md): Documents that nodes build from their results — conclusion first, evidence underneath.
- [Flow editor](https://docs.ovalbee.com/build/editor.md): Everything you can do on the canvas by hand, and the shortcuts that make it fast.
- [Build with an AI agent](https://docs.ovalbee.com/build/agent.md): Describe the workflow in plain words and let an agent assemble, explain or repair the flow.
- [Templates and sharing](https://docs.ovalbee.com/build/templates.md): Start from a proven flow, save your own, move flows between workspaces and give people access.
- [Use cases](https://docs.ovalbee.com/use-cases/overview.md): Common computer vision jobs, the flow shape that solves each one, and a template to start from.
- [Bring data into OvalBee](https://docs.ovalbee.com/use-cases/bring-data-in.md): Upload your own images and videos, pull a public dataset, or import a Supervisely project.
- [Organize labeling](https://docs.ovalbee.com/use-cases/organize-labeling.md): Turn a pool of images into reviewed annotations, with every image's state visible on the canvas.
- [Check and clean a dataset](https://docs.ovalbee.com/use-cases/check-and-clean-data.md): Find what's wrong with a dataset before training does it for you, and fix it with rules you can rerun.
- [Try a pretrained detector](https://docs.ovalbee.com/use-cases/try-a-pretrained-detector.md): See what an off-the-shelf model finds on your own images before you spend anything on labels.
- [Train a detector](https://docs.ovalbee.com/use-cases/train-a-detector.md): Split a labeled dataset, fine-tune a detector on it and see whether training actually converged.
- [Evaluate and compare models](https://docs.ovalbee.com/use-cases/evaluate-and-compare.md): Score a detector against ground truth, and choose between several with the same evaluation.
- [Administration](https://docs.ovalbee.com/administration/overview.md): Setting OvalBee up for a team — where it runs, who can do what, and what an administrator owns.
- [Members and roles](https://docs.ovalbee.com/administration/members-and-roles.md): Who belongs to a workspace, what each role can do, and how to give someone access to a single flow.
- [Secrets and API tokens](https://docs.ovalbee.com/administration/secrets-and-tokens.md): Credentials that nodes use to reach other systems, and tokens that let scripts and agents act for you.
- [Sign-in and accounts](https://docs.ovalbee.com/administration/sign-in-and-accounts.md): Single sign-on, how accounts are created, passwords and account deletion.
- [Audit log](https://docs.ovalbee.com/administration/audit-log.md): An append-only record of who signed in, who changed access, and who created or deleted what.
- [Compute](https://docs.ovalbee.com/administration/compute.md): Where node runs and deployments execute, and how the ones that need a GPU get one.
- [Platform configuration](https://docs.ovalbee.com/administration/platform-configuration.md): Every instance-level setting, tab by tab.
- [Self-hosted installation](https://docs.ovalbee.com/administration/self-hosting.md): What an OvalBee installation on your own infrastructure consists of, and how to prepare for it.
- [Glossary](https://docs.ovalbee.com/resources/glossary.md): The terms used across OvalBee and these docs, in one place.
- [Node Library](https://docs.ovalbee.com/node-library/nodes/overview.md): Every node you can place on the canvas, what it does and what it needs.
- [Import drag & drop](https://docs.ovalbee.com/node-library/nodes/import-ops-import-drag-and-drop.md): Import images and videos into a dataset by dropping them onto the node. Each import adds a new version of the destination dataset, and you choose what happens when a file name is already taken.
- [Import from Dataset Ninja](https://docs.ovalbee.com/node-library/nodes/import-ops-import-from-dataset-ninja.md): Import a full dataset or a sample from Dataset Ninja, including its images, annotations, splits and image-level labels.
- [Import from cloud storage](https://docs.ovalbee.com/node-library/nodes/import-ops-import-from-cloud-storage.md): Import images or videos with their annotations from an S3-compatible bucket. The annotation format is detected automatically.
- [Export to cloud storage](https://docs.ovalbee.com/node-library/nodes/export-ops-export-to-cloud-storage.md): Export a dataset's images, with their annotations, to an S3-compatible bucket.
- [Dataset](https://docs.ovalbee.com/node-library/nodes/dataset-ops-dataset.md): Points the flow at a dataset and shows its assets and annotations on the node card. Other nodes read the dataset from this node's input.
- [Annotation quality](https://docs.ovalbee.com/node-library/nodes/dataset-ops-annotation-quality.md): Check the annotations of an image dataset and report coverage, class balance, and issues such as duplicate images, out-of-frame labels, and split leakage.
- [Dataset insights](https://docs.ovalbee.com/node-library/nodes/dataset-ops-dataset-insights.md): Turns the dataset's statistics into findings written in plain language, each one backed by a chart in the report.
- [Find annotation gaps](https://docs.ovalbee.com/node-library/nodes/dataset-ops-find-annotation-gaps.md): Find missing annotations and possible class confusion using confident external predictions. Prepare missing-object candidates and report class conflicts for human review. Use predictions from an external model, not one trained on the same dataset.
- [Image quality](https://docs.ovalbee.com/node-library/nodes/dataset-ops-image-quality.md): Measure sharpness, exposure and broken files from the pixels alone, for every image including the ones nobody labelled. The first run downloads every image of the dataset. What it measured is kept per image, so later runs download only new files.
- [Copy](https://docs.ovalbee.com/node-library/nodes/dataset-ops-copy.md): Adds images and videos from one OvalBee dataset to another without duplicating the media files.
- [Move](https://docs.ovalbee.com/node-library/nodes/dataset-ops-move.md): Moves images and videos, with their annotations, from one OvalBee dataset to another and removes them from the source.
- [Merge datasets](https://docs.ovalbee.com/node-library/nodes/dataset-ops-merge-datasets.md): Merge two OvalBee datasets into one, choosing which images to keep and how to source and resolve their annotations.
- [Create ontology](https://docs.ovalbee.com/node-library/nodes/dataset-ops-create-ontology.md): Defines the ontology the rest of the flow works in. Pick one that already exists, or create one right here and fill it with classes without leaving the graph.
- [Remove assets](https://docs.ovalbee.com/node-library/nodes/asset-ops-remove-assets.md): Removes images and videos, with their annotations, from an OvalBee dataset - all of them, or only the ones a second dataset also holds.
- [Transform annotations](https://docs.ovalbee.com/node-library/nodes/asset-ops-transform-annotations.md): Writes a dataset's images with transformed annotations into another dataset: keeps selected classes, keeps a single label shape, and renames classes.

- [Filter images](https://docs.ovalbee.com/node-library/nodes/filter-filter-images.md): Removes or keeps whole images of an OvalBee dataset by labels count, asset tags, classes, image size, and object-level conditions (area share, box aspect, geometry, object tag), combined with And or Or, then adds the selected images to a destination dataset. It acts on whole images - an image matches the object conditions when one of its objects satisfies all of them at once, and the image moves to the destination with all of its labels. To clean up individual labels instead, use Filter annotations.
- [Filter annotations](https://docs.ovalbee.com/node-library/nodes/filter-filter-annotations.md): Removes or keeps individual labels of an OvalBee dataset by class, area share, box aspect, geometry, and object tag. It acts on each matching label and images stay in place, writing the images with their filtered annotations to a destination dataset. To move whole images instead, use Filter images.
- [Filter predictions](https://docs.ovalbee.com/node-library/nodes/filter-filter-predictions.md): Selects a subset of a dataset from the confidence of its prediction annotations. For classification, keeps the assets whose predicted confidence falls inside a [Conf min, Conf max] band and samples a budget across the predicted classes, e.g. a mid-confidence band of 'hard' images for active learning. For object detection, scores every asset from its box confidences and keeps the highest-scoring ones above a floor, e.g. to select frames for pseudo-label training.
- [Random sample](https://docs.ovalbee.com/node-library/nodes/sample-random-sample.md): Draws a random sample of images from a source dataset and adds them, with their annotations, to a destination dataset. Images drawn in earlier runs are skipped, so each run returns new ones. Without a destination, the node's own output dataset holds only the latest sample.
- [Train/validation split](https://docs.ovalbee.com/node-library/nodes/sample-train-validation-split.md): Splits a dataset's images, with their annotations, into train and validation datasets and an optional test dataset, either copying them or moving them out of the source.
- [Review suspected label errors](https://docs.ovalbee.com/node-library/nodes/sample-review-suspected-label-errors.md): Collects the images an object detection evaluation report flagged as probable ground-truth label mistakes into a dataset, ready to send for re-labeling.
- [Labeling gate](https://docs.ovalbee.com/node-library/nodes/label-labeling-gate.md): Run rule-based checks on freshly submitted annotations and route each image to a passed or failed dataset before human review.
- [Labeling](https://docs.ovalbee.com/node-library/nodes/label-labeling.md): Hand images or videos from a source dataset to labelers in small batches. Each outcome status becomes a button in the annotation tool that moves the submitted asset and its annotation to that status's dataset.
- [Label train, validation and test](https://docs.ovalbee.com/node-library/nodes/label-labeling-splits.md): Hand a train, a validation and a test set to labelers from one queue, split by split, and keep each split's labeled images in a dataset of its own.
- [Review](https://docs.ovalbee.com/node-library/nodes/label-review.md): Hand annotated images or videos from a source dataset to reviewers in small batches. The annotation tool opens read-only, and each outcome status becomes a verdict button that moves the asset and its annotation to that status's dataset.
- [Train YOLO](https://docs.ovalbee.com/node-library/nodes/models-yolo-train-yolo.md): Train a YOLO object detection or instance segmentation model directly on OvalBee datasets.
- [Train DEIM](https://docs.ovalbee.com/node-library/nodes/models-deim-train-deim.md): Train a DEIM object detection model directly on OvalBee datasets.
- [Train EdgeCrafter](https://docs.ovalbee.com/node-library/nodes/models-edgecrafter-train-edgecrafter.md): Train an EdgeCrafter object detection or instance segmentation model directly from OvalBee datasets.
- [Predict YOLO](https://docs.ovalbee.com/node-library/nodes/models-yolo-predict-yolo.md): Run a YOLO detection or instance segmentation model on a dataset and write the images with their predictions to the output dataset.
- [Predict DEIM](https://docs.ovalbee.com/node-library/nodes/models-deim-predict-deim.md): Run DEIM object detection inference on a dataset and write the images and the DEIM prediction annotations into the output dataset.
- [Predict EdgeCrafter](https://docs.ovalbee.com/node-library/nodes/models-edgecrafter-predict-edgecrafter.md): Run EdgeCrafter inference (object detection or instance segmentation) on a dataset and write the images and the EdgeCrafter prediction annotations into the output dataset.
- [Prelabel with live training](https://docs.ovalbee.com/node-library/nodes/models-general-prelabel-with-live-training.md): Pre-label a dataset's images with the weights a running live training session has right now, into a new dataset or in place for labeling, without stopping the session.
- [Evaluate object detection](https://docs.ovalbee.com/node-library/nodes/evaluate-evaluate-object-detection.md): Calculate object detection metrics by comparing predictions against ground truth, both supplied as separate datasets with their annotations.
- [Compare object detectors](https://docs.ovalbee.com/node-library/nodes/evaluate-compare-object-detectors.md): Compare detector evaluations on the same ground truth, or two annotation versions using the same model predictions and evaluation settings.
- [Active learning curve](https://docs.ovalbee.com/node-library/nodes/evaluate-active-learning-curve.md): Record one point per active learning cycle and chart mAP on a fixed evaluation set against the number of labeled training images.
- [Recommend training config](https://docs.ovalbee.com/node-library/nodes/evaluate-recommend-training-config.md): Read an object detection evaluation report and propose training settings for the next run.
- [Evaluate prelabeling efficiency](https://docs.ovalbee.com/node-library/nodes/evaluate-evaluate-prelabeling-efficiency.md): Evaluate how well an object detection model works as a labeling assistant. Reports an assistance score and a breakdown of correct, shifted, class-confused, false positive, and false negative predictions.
- [Export YOLO](https://docs.ovalbee.com/node-library/nodes/models-yolo-export-yolo.md): Convert a YOLO checkpoint to PyTorch, ONNX and TensorRT and attach every format to the model it came from.
- [Export DEIM](https://docs.ovalbee.com/node-library/nodes/models-deim-export-deim.md): Convert a trained DEIM checkpoint to PyTorch, ONNX and TensorRT and attach every format to the model it came from.
- [Embedding anomalies](https://docs.ovalbee.com/node-library/nodes/models-clip-embedding-anomalies.md): Score indexed images by how far they sit from their nearest neighbours and split the dataset into anomalies and typical images.
- [Label diagnostics](https://docs.ovalbee.com/node-library/nodes/models-clip-label-diagnostics.md): Find likely object-label outliers and directed class confusion in an annotated dataset.
- [Embedding map](https://docs.ovalbee.com/node-library/nodes/models-clip-embedding-map.md): Cluster a dataset by visual similarity and name each cluster with a vision-language model, either as a 2D map of the whole dataset or as rare subtypes inside one image-level tag.
- [Similar pairs](https://docs.ovalbee.com/node-library/nodes/models-clip-similar-pairs.md): Find all similar image pairs inside a dataset or across a reference dataset.
- [AI index embeddings](https://docs.ovalbee.com/node-library/nodes/models-clip-ai-index-embeddings.md): Embed a dataset's images and annotated objects into the workspace index, and keep it current as the dataset changes.
- [Upload model](https://docs.ovalbee.com/node-library/nodes/models-general-upload-model.md): Upload a local .pt or .pth checkpoint and register it as a new version in the workspace model registry.
- [Run flow](https://docs.ovalbee.com/node-library/nodes/flow-control-run-flow.md): Start the chain of nodes connected to this node. Performs no work itself.
- [Wait](https://docs.ovalbee.com/node-library/nodes/flow-control-wait.md): Wait for all incoming execution branches to complete before continuing. Place at merge points in parallel flows.
- [Stop flow](https://docs.ovalbee.com/node-library/nodes/flow-control-stop-flow.md): End this path of the flow run here. Nodes connected after it do not start, and other branches keep running.
- [Import images](https://docs.ovalbee.com/node-library/nodes/supervisely-import-images.md): Imports images from Supervisely into OvalBee and adds them to the destination dataset.
- [Labeling queue manager](https://docs.ovalbee.com/node-library/nodes/supervisely-labeling-queue-manager.md): Runs one Supervisely labeling queue for a whole step and shows where every image stands as OvalBee stage datasets.
- [Labeling jobs manager](https://docs.ovalbee.com/node-library/nodes/supervisely-labeling-jobs-manager.md): Hands batches out as ordinary Supervisely jobs, one batch per person, and re-issues rejected work as a new job.
- [Labeling stats](https://docs.ovalbee.com/node-library/nodes/supervisely-labeling-stats.md): Running totals of the labeling side of a Supervisely process: how many people label and how many images they have submitted.
- [Review stats](https://docs.ovalbee.com/node-library/nodes/supervisely-review-stats.md): Running totals of the review side of a Supervisely process: who reviews, how much went through and how often work came back.
- [Changelog](https://docs.ovalbee.com/changelog/llms.txt)