Inside RemoteCode Bots
Give recurring work a dedicated Bot.
Create AI teammates with their own instructions, memory and skills. Bring them back to the same job without rebuilding the setup every time.
Get RemoteCode for MacToken Saver and remote control are subscription features.
What should I put in a Bot’s instructions?
Start with a specific responsibility and a result you can check. A review Bot might inspect a set of changes and return the affected file, the reason for concern and the evidence it found. Tell it which project to use, what is outside the task and whether you expect analysis or edits. These are instructions you write, not automatic guarantees. Keep the role narrow enough that you can judge the answer. You can choose the provider and model for each task; the Bot is a reusable working setup, not a replacement for that provider or its account.
How do memory and skills differ?
Instructions describe the role. Memory keeps reusable context in Markdown files, so a recurring job can refer to prior decisions without requiring you to repeat the whole setup. Skills provide task-specific guidance; they are indexed and loaded when requested. Use memory for durable facts such as project conventions, and skills for a procedure the Bot needs to follow. Review both when a project changes. A stored instruction can become outdated, and having a skill available does not mean every task will load it or follow it successfully. Check the result against the current project.
How should I start a scheduled task?
First try the prompt in a conversation and inspect the result. Describe the material to read, the output you need and a clear stopping point. Once that task is useful, choose its frequency and time zone for recurring work. For example, a weekday review can ask for yesterday’s changes and questions to investigate before release. Treat that output as a review aid. Your Mac must remain powered on and available, with access to the selected provider and the project. Scheduling is not a cloud execution service, and a saved schedule alone does not prove that a run completed.
When should I mention another Bot?
Use an @handle when another defined role has useful expertise for the current task. You might ask a writing Bot to consult a technical reviewer about a claim before revising the wording. The active agent decides whether to make the consultation and writes a focused request. A quoted mention or a request not to involve a Bot does not trigger delegation. The reply stays connected to the consultation, which lets you review the context of the answer. Mentioning a specialist does not establish that its conclusion is correct; compare its evidence with the original task before acting.
What changes when I continue from my iPhone?
The phone gives you remote access to your Bots and their conversations. You can inspect progress and send the next instruction while execution continues on the Mac. It does not move your project to the phone or make a sleeping, disconnected Mac available. Before relying on a remote session, check that the Mac is reachable and that the relevant provider is usable. Remote access is part of the paid mobile offering shown on this page. Your AI provider remains separate: RemoteCode does not supply extra provider quota or remove provider-side limits.
How can I evaluate the result of recurring work?
Give the first task a small scope with an observable outcome. Ask the Bot to distinguish what it inspected, what it changed and what it could not verify. Read the resulting conversation before relying on it for a release, a publication or another consequential action. If it missed the goal, refine the instructions or the task prompt before increasing the frequency. For tasks with large command output, the Token Saver guide explains how context compression and recovery work. Savings depend on the workload; shorter context is not evidence that the Bot’s final answer is complete or accurate.
How RemoteCode saves tokens