YC-Incubated Reason Machines Launches Open Source Reason Network for Cross-Computer Coding Agents Communication
YC-incubated Reason Machines has launched the open-source Reason Network, allowing coding agents like Claude Code, Codex, Cursor, and OpenCode on different computers to communicate directly.
The system establishes an address for each agent session, displaying the associated developer, working repository, branch, and current task status, supporting one agent to send messages to another agent on a colleague's computer.
For example, Codex in web development can directly inquire whether the backend-related Claude Code session interface has changed, eliminating the need for humans to relay information across multiple chat windows.
The underlying connection is established via iroh for P2P communication, and if direct connection fails, an encrypted relay is used. Once the message reaches the target computer, it is injected into the corresponding open session via session Socket, CLI, or Hook mechanism.
Running agents can receive messages directly without needing to start a new task for communication, and the team directory updates the busy status and work content of each session in real-time.
From a market perspective, this tool reduces the manual transfer costs of multi-agent collaboration, shifting resources from human coordination to direct communication between agents, benefiting development teams and open-source tools, while solutions relying on centralized coordination platforms face substitution pressure.
Additional information indicates that the network engine is open-sourced under Apache 2.0, with each computer running a lightweight daemon that supports direct LAN connections and internet tunneling, with messages transmitted only among team members.
Source: Public Information
ABAB AI Insight
As a YC-incubated company, Reason Machines focuses on building autonomous software engineering and coding agent tools, having early on launched a cloud-based agent that connects to GitHub and submits evidence-backed pull requests, with a historical trajectory expanding from single-agent task execution to multi-agent collaboration.
Their capital path reflects the open-sourcing of the network layer, attracting developers to use local agents for interconnection through free P2P connections, while retaining paid access to cloud sessions, expanding resources from model invocation to cross-device messaging and team directory services.
Similar cases include other multi-agent messaging protocols and local daemon projects, as well as the early evolution of distributed collaboration tools from centralized servers to peer-to-peer systems. The current coding agent ecosystem is transitioning from single-session tools to real-time team collaboration.
This essentially represents a restructuring of the industry chain: agent communication shifts from human intermediaries to direct machine addressing and message injection, reducing coordination delays through session-level addressing and P2P transmission, allowing multi-agent workflows to progress in parallel without continuous human intervention.
ABAB News · Cognitive Laws
Direct connections between agents reduce human intermediaries, with efficiency depending on addressing accuracy.
Open-source network layers expand adoption, while cloud sessions retain monetization entry points.
Existing sessions can receive messages without needing to restart tasks for communication.