# Federation over Text: Insight Sharing for Multi-Agent Reasoning
> Federation over Text (FoT) is a multi-agent AI framework where LLM agents share reusable reasoning insights as natural language instead of gradients. Agents build a shared insight library for collective reasoning, improving accuracy and reducing reasoning tokens across math, cross-domain collaboration, and research insight tasks.
Type: paper
Venue: ICLR 2026 Workshops MALGAI and RIS
Date: 2026
Authors: Dixi Yao, Tahseen Rabbani, Tian Li
## Direct Links
- [Canonical topic page](https://dixiyao.github.io/topics/federation-over-text/)
- [Project page](https://dixiyao.github.io/fot/)
- [Paper](https://arxiv.org/abs/2604.16778)
- [PDF](https://arxiv.org/pdf/2604.16778)
- [Code](https://github.com/dixiyao/FoTClaw)
- [Markdown alternate](https://dixiyao.github.io/fot/index.html.md)
## Search Queries and Aliases
- FoT
- Federation over Text
- multi-agent AI
- multi-agent reasoning
- multi-agent system
- mult-agent
- federation agent
- federation plus agent
- federated agents
- LLM agent federation
- FoTClaw
- semantic federation
- insight sharing
## Search Intents
- multi-agent agents sharing reasoning insights
- federation over text paper
- FoTClaw code
- federated learning inspired agent collaboration
- LLM agents with shared insight library