Understand
Ground decisions in long, multimodal, evidence-rich information.
A research program for executable work intelligence
Office Intelligence
Building simulated organizations where agents understand information, execute real workflows, and learn from interaction.
Read our visionOur vision
Make knowledge work executable, observable, and learnable.
At the core of Office Intelligence are executable simulated organizations: persistent environments where agents take on roles, use tools, coordinate under rules, and work toward shared goals over time. These organizations form a shared environment layer for dynamic evaluation, synthetic data generation, and reinforcement-learning-based post-training.
Act
Use tools, coordinate roles, and complete real workflows over time.
Improve
Turn interaction traces into evaluations, training data, and RL signals.
Current research systems
Building the environment layer for evidence, interaction, and learning.
Benchmark · 2026
01 / BENCHMARK
XL-DocBench
Evidence-grounded reasoning across hundreds or thousands of pages, with 1,519 human-verified questions and exact page-level support.
- 2,303
- maximum context pages
- 72.6%
- multi-page evidence
- 194
- human experts
Agent system · 2026
02 / AGENT SYSTEM
DocAtlas
Long-document understanding as mutable-state interaction through Search, Read, Note, Review, and end-to-end reinforcement learning.
- 71.4%
- MMLongBench-Doc
- 63.7%
- RL-trained 4B policy
- 4
- stateful tools