OneLLMONELLM · Guides
AI agents on iPhone: choose by the work you need done.
Before choosing an AI agent for your iPhone, separate sending instructions from executing the work. The difference determines network access, file permissions, model costs and whether a computer must stay on.
Local AI, remote coding and office agents
Swipe horizontally to view the full table.
| Task | Where it runs | What to verify |
|---|---|---|
| Offline AI conversation | A downloaded local model performs inference on the phone. | The model is downloaded and fits the device’s memory and storage. Network tools are separate from offline inference. |
| Operate OpenCode or DS Harness remotely | The phone connects to a coding service on the computer, where the project and execution environment reside. | The computer is on, the service is reachable, the model account works and the pairing module matches. |
| Work on office files | Files are handled in the environment supported by the selected product; a mobile interface alone does not determine it. | Release availability, accessible files, required output formats and a way to review results. |
Choose by your first task
I want to draft or summarize supplied text without a network
Start with OneLLM’s local model workflow. Download a suitable model beforehand and try it on the intended device. Larger models need more resources: the optional 35B MoE model needs about 16.6 GB of storage, with at least 12 GB of memory recommended. Do not assume every iPhone is suitable for the largest model.
I want to continue a coding conversation away from my desk
If you already use OpenCode, follow the OpenCode mobile guide to connect OneLLM. The computer continues running the service, with the phone as the conversation interface. For DS Harness, use its matching module rather than an OpenCode pairing code.
I want to work with spreadsheets, reports and computer files
Compare file access, output formats and current availability. For computer agents, read the DeepSeek Harness and OpenCode comparison. OneLLM Office remains a preview on this site, without a desktop installer offered here; the mobile app and connector are separate from a released Office desktop product.
A useful remote coding handoff
A phone screen works well for setting a clear goal and reviewing a summary. Define the scope so you do not return to your desk to discover changes in the wrong project.
Check why the login page in the current project scrolls horizontally on narrow screens. First identify the files involved and propose a fix. Confirm the project directory before editing. Report the changes, validation and anything untested. Do not deploy.
- Scope:identify the project, file or page and actions outside the task.
- Outcome:define completion, such as no horizontal overflow at a specified screen width.
- Review:request the changes and validation evidence. Return to the computer to inspect the diff and run results when needed.
- Connection:test from the network you intend to use and ensure computer sleep will not make the service unreachable.
Three limits worth checking
A mobile interface does not mean all work runs on the phone
Remote OpenCode conversations depend on the computer service. Downloaded OneLLM local models can perform inference offline. Cloud models, web searches and remote computer access still need the corresponding network connections.
One client does not imply support for every agent
This guide describes the OpenCode and DS Harness connections offered on this site. It does not imply that Claude Code, Codex or other agents can pair with OneLLM using the same connection code.
Connection does not include free model usage or complete platform validation
Computer services use their own model accounts with separate charges. The connectors are previews; see platform testing status. Try a small task with your devices and network before relying on the setup for regular work.