What each part does
This is an operating procedure you can use with PeopleBot, not a built-in Termius integration or an automatic feature of the public package.
1. Ask your assistant to prepare the connection
Start with Termius installed on your phone, an authorized remote connection, and an agent CLI that already works on the target computer. Your administrator supplies connection and authentication settings privately. A PeopleBot project assistant with access to that computer can inspect the environment, prepare a launcher, and give you the exact Termius settings and snippet to enter. Ask:Help me reach this project agent from Termius on my phone. Inspect the operating system, shell, installed agent CLI, and project folder. Prepare and test a named persistent session that reconnects to the existing agent before creating anything new. Give me a short connection label and the exact launcher command for this environment. Keep credentials and connection details out of public documentation. Tell me what you verified and what still needs a phone test.The launcher should enter the correct project, use the installed agent’s supported startup or resume procedure, and attach to an existing session when present. We have used tmux on Linux and psmux on Windows for persistent terminal sessions. The assistant must check the tool and version on your computer; commands are not interchangeable. On Windows, have the assistant test the actual shell and executable entry point. If a launcher is blocked by a script policy, diagnose that specific problem rather than broadly disabling protections.
2. Give every agent a recognizable label
In Termius, create or select the host using the private connection settings supplied for you. Use labels that describe the work:
These are examples, not a required PeopleBot configuration. Your setup might have several sessions on one computer or agents on different computers.
Connect and confirm the expected project and agent before assigning work. Keep addresses, usernames, authentication material, and screenshots containing connection details private.
3. Save the tested launcher as a snippet
A Termius snippet saves a shell command so you do not have to type it repeatedly on a phone. On mobile, open Vaults → Snippets, tap +, choose New Snippet, and enter a name and the tested launcher in Script. Save it. A name such asconnect-demo-reviewer makes the target easy to recognize.
From a connected terminal, open the additional keyboard’s four-square menu and choose Snippets. On iPad, use the right-side panel. Run the launcher at a shell prompt, before entering the agent conversation. Sending it while the agent is accepting messages would give the command to the agent instead.
After testing, you can select a snippet under the host’s Startup Command setting. Use a launcher that attaches to the existing named session instead of starting a duplicate agent. Select only the intended target; do not broadcast a task to multiple hosts.
See Termius’s snippet instructions for current controls and behavior. Feature availability can depend on your plan and app version.
4. Use the phone as your agent console
Once the agent is accepting input, type ordinary requests. You do not need to begin with a complicated keyboard shortcut. Check where you are:Tell me which project and agent Instance this conversation belongs to, what you are working on, and whether anything needs my decision.Assign a bounded task:
Read the current task and saved project context. Complete the assigned documentation change, check the result, and save a report with the changed files, evidence, and anything unresolved.Review before continuing:
Show me the saved result and the checks that support it. Separate completed work from planned or unverified work.Prepare to leave:
Save a concise handoff with the current task, decisions, exact saved work, checks, and next step. Tell me where it is stored.Switch to another labeled session when you need another agent. Typing through Termius is your direct conversation with that agent; any agent-to-agent coordination is a separate part of the project workflow. Follow Run a project with your assistant for task structure and Check whether work is finished for reviewing the answer.
5. Test reconnecting without restarting the work
Before relying on the setup:- Ask the agent to identify its project and current conversation.
- Send a small status request and confirm the response.
- Disconnect the phone without exiting the agent process.
- Reconnect using the saved launcher.
- Confirm you returned to the same session and did not start another agent.
When something looks wrong
Terminal scrollback is convenient, but it is not your durable project record or a verified backup. Save important results and use Continue in a new chat and Verify your backup when needed.