Annotate anything — text, agents, and beyond
Potato is an open-source platform for annotating text, audio, images, and qualitative data — from classification and span labeling to codebook-driven coding. It's also the first annotation tool purpose-built for evaluating AI agents: import traces from any framework, visualize agent actions interactively, and configure it all in YAML.
pip install potato-annotation# Create your config.yamlpotato start config.yaml# Open localhost:8000
Everything you need for annotation
Powerful features designed for researchers and teams.
Agent Evaluation
Annotate multi-agent teams on a clickable interaction graph, evaluate computer-use, voice, and video agents, calibrate LLM judges against human labels, edit trajectories into SFT and DPO training data, and rank models in an arena — all via YAML.
30+ Annotation Types
Radio, multiselect, likert, slider, text, span, best-worst scaling, pairwise, number, multirate, video, image, audio, bounding box, polygon, event, and taxonomy annotation, plus qualitative coding (QDA Mode) with a living codebook, in-vivo codes, memos, and cases.
AI-Powered
LLM integration with OpenAI, Claude, Gemini for intelligent hints, keyword highlighting, label suggestions, MACE competence estimation, option highlighting, and diversity ordering.
Multimedia Support
Annotate audio with waveforms, images with bounding boxes and polygons, and video with playback controls.
Active Learning
5 query strategies — uncertainty sampling, diversity-based selection, BADGE, BALD, and hybrid ensemble — plus LLM cold-start for intelligent instance selection before any labels exist.
Zero Code Setup
Configure everything in YAML. No programming required to create sophisticated annotation interfaces.
Built for every domain
Text, images, audio, or video — Potato has you covered.
350+
Browse 350+ annotation designs
Ready-to-use configurations from the community.
Ready to start annotating?
Join researchers and teams worldwide using Potato.