# Toloka

> Toloka builds expert human data and real-world data pipelines for frontier AI: agent training and evaluation, LLM/VLM fine-tuning data, coding data, AI-safety red-teaming, and Physical AI collection. Physical AI spans deployed egocentric/exocentric capture, 25,000+ hours of off-the-shelf egocentric video, and the open-source HomER egocentric robotics dataset. Toloka is part of Toloka AI BV (Amsterdam, Netherlands) and serves clients including Amazon, Microsoft, Anthropic, and Shopify.

## Physical AI
- [Physical AI data](https://toloka.ai/physical-ai): Egocentric and exocentric real-world capture (deployed); teleoperation and UMI (in development); custom collection design, annotation, and managed QA.
- [Robotics model data](https://toloka.ai/robotics): Training data for VLA and manipulation models.
- [HomER dataset](https://toloka.ai/blog/homer-v2-a-larger-more-diverse-egocentric-dataset-for-robotics-research/): Open egocentric robotics dataset — ~100 hours / 765 first-person videos in v2; 25K+ hours available off the shelf. Data: https://huggingface.co/datasets/toloka/HomER

## Data solutions
- [AI agent data](https://toloka.ai/agent-data): Trajectories, tool-use data, RL environments, MCP replicas, and agent evaluation.
- [LLM & VLM data](https://toloka.ai/large-language-models): SFT, preference, reasoning-chain, and domain-expert data.
- [Coding data](https://toloka.ai/coding-data): Training and evaluation data for coding models.
- [AI evaluation](https://toloka.ai/evaluation): Human and model-based evaluation for LLMs and agents.
- [AI safety & red teaming](https://toloka.ai/ai-safety): Adversarial testing and safety datasets.
- [Creative AI data](https://toloka.ai/creative-data): Data for creative and multimodal generation.
- [Hybrid human-synthetic data](https://toloka.ai/hybrid-data): Blended human + synthetic data pipelines.

## Research & benchmarks
- [Toloka Arena](https://toloka.ai/arena): Independent evaluation of agentic intelligence.
- [Research](https://toloka.ai/research): Published research on data methods, evaluation, and responsible AI.

## Optional
- [Case studies](https://toloka.ai/success-stories): How companies use Toloka for post-training and evaluation.
- [Blog](https://toloka.ai/blog/): Articles on data, evaluation, and Physical AI.