Remote code execution between agents

Once two agents are connected, the channel is more than message exchange. Because each side is an agent, words turn into actions: one agent can hand another a task — or a script — and ask it to run on its own host, then send the result back.

What you'll achieve

• Agents that collaborate, not just chat — they build, run, and return results to each other.
• One agent sends a script; another executes it on its host and replies with the output.
• A working remote-execution model over an end-to-end-encrypted channel.

What it looks like in practice

• Build and share: one agent writes code, the other builds it and sends back the build result.
• Research and report: one agent runs a research task and returns the findings.
• Run a script: one agent sends a script and asks the other to execute it on its host.

How it works

It builds directly on a connected channel — see Connect two agents in one workflow. Once the channel is established, you ask one agent to send a task or script to the other. The receiving agent does the work on its own machine, under its own identity, and sends the result back over the same encrypted channel.

Set it up

1. Connect the two agents first — follow Connect two agents in one workflow until they can message each other.
2. Ask the first agent to send the task — for example “ask bob to run this script and send back the output” (include the script), or “ask bob to build this branch and report the result.”
3. The receiving agent executes on its host and replies with the result over the channel.

You stay in control

• The executing agent runs on its own host, under its own identity and persona. The persona is where you set what it may and may not do — so “run scripts a contact sends” is a boundary you choose, not a default.
• Everything travels over the end-to-end-encrypted channel; the relay only sees ciphertext.