Overview

Pictograph turns directories of images into small, task-specific vision models (SAM3 drafts the labels, you review them) and makes those models tools your agent can call. One REST API drives three surfaces: the pictograph CLI (how a coding agent drives Pictograph), a typed Python SDK, and an agent tool registry.

from pictograph import Client, get_model, DetectionModel, DetectionResult

client = Client()  # reads PICTOGRAPH_API_KEY

model: DetectionModel = get_model(
    name="Road Signs",
    task="object_detection",
    client=client,
)
result: DetectionResult = model.predict(
    image="photo.jpg",
)

for prediction in result.predictions:
    print(prediction.name, round(prediction.confidence, 2), prediction.bounding_box)

What you can do

• Hand the loop to your coding agent - install the CLI and the pictograph-cv skill, and Claude Code or Codex drafts labels for you to review, then exports, trains, and deploys the model. See Agents.
• Upload directories of images; subdirectories become virtual paths.
• Auto-annotate with SAM3 - point, box, or text prompts, single image or async batch, or pre-label with a model you trained.
• Review the drafts before you export - filter to the least certain, then approve or send back each image.
• Train detection (YOLOX, RF-DETR), segmentation (instance and semantic), keypoint, and classification models on managed GPUs, then evaluate them.
• Deploy a model as one authenticated /predict endpoint, or run a saved workflow over images and video.
• Export to 12 formats - COCO, YOLO (incl. OBB and pose), DOTA, Darwin V7, Pascal VOC, CVAT, Datumaro, LabelMe, CSV, or Pictograph JSON.
• Import existing datasets from V7 (Darwin) or Roboflow, or from COCO, YOLO, and Pascal VOC files.
• Search by visual similarity or auto-generated content tags.
• Run locally with Pictograph’s own tooling - feed a PyTorch training loop, run trained models on your own hardware, and draw annotations, no server round-trip.
• Build your own agent on the tool registry - Claude Agent SDK, openai-agents, or any framework that speaks JSON Schema.

Map of the docs

Every page has a “Copy as Markdown” button and an .md mirror for agents to consume directly. Documented methods link to the code that implements them, pinned to the released tag.

For agents browsing this site

• Site index: /docs/llms.txt
• Full doc bundle (one file): /docs/llms-full.txt
• Tool registry (JSON Schema): /api/v1/developer/tools.json
• SDK source: pictograph-io/pictograph-sdk