Introduction
Mobile AI agents are moving from cloud‑only to truly on‑device experiences, and developers need a reliable way to test and ship them. dsh‑Go is defined as an unofficial Android client that lets you run DeepSeek Harness either locally on your phone or remotely via SSH, while routing inference to any OpenAI‑compatible API you configure.
What is DeepSeek Harness and How dsh‑Go Extends It
DeepSeek Harness (dsh) is an open‑source AI coding‑agent framework that provides a CLI, plugin ecosystem, and workflow engine. The official docs describe installing it with npx @deepseek-ai/dsh web and configuring API keys, workspaces, and agents. dsh‑Go packages a minimal Ubuntu + Node.js environment inside an Android app, allowing the same harness to run on a phone or act as a remote controller for a PC/server.
Installing and Configuring dsh‑Go
- Download the latest APK from the GitHub releases (v1.1.0 released 2026‑11‑15).
- Install the app on a device running Android 8.0 or higher.
- On first launch, grant storage and network permissions; the app will unpack the Ubuntu image.
- Open the Settings screen and add your OpenAI‑compatible API key (e.g., OpenAI, Azure, or a self‑hosted endpoint).
- Choose a workspace directory or sync with a cloud folder to store agent scripts.
Choosing the Right Runtime Mode
dsh‑Go supports three primary modes:
- Local Mode: The harness runs entirely on the phone using the bundled Ubuntu environment. Ideal for quick prototyping and offline demos.
- Remote Mode: The app connects over SSH to a PC or server where dsh is installed. Use this when you need more compute power or access to GPUs.
- Hybrid Mode: Core agent logic runs locally while heavy model inference is delegated to the remote server, balancing latency and resource usage.
Select the mode that matches your performance needs and security policies before launching an agent session.
Wiring an OpenAI‑Compatible API Securely
Regardless of the runtime mode, inference requests are sent to the API endpoint you configured. Follow these best practices:
- Store the API key in the app’s encrypted keystore, never in plain text.
- Use HTTPS for all endpoint calls; validate SSL certificates.
- Enable rate‑limiting or request throttling in the harness config to avoid accidental over‑usage.
- For remote mode, ensure the SSH tunnel uses key‑based authentication and disables password login.
Shipping Safely: Testing, Monitoring, and Updates
Before releasing an app that embeds dsh‑Go, run a full QA cycle:
- Validate that the harness starts within 30 seconds on target devices.
- Simulate network loss and verify graceful fallback to local mode.
- Monitor API response times and error rates using a lightweight logging library.
When updates are available, the app can download a new Ubuntu image and restart the harness without requiring a full reinstall, ensuring a smooth user experience.
Frequently Asked Questions
Can dsh‑Go run on iOS devices?
No. dsh‑Go is built as an Android client because it relies on a Linux container that is not supported on iOS.
Do I need a DeepSeek account to use dsh‑Go?
You only need an API key for the inference provider you choose. The harness itself is open source and does not require a DeepSeek subscription.
How does dsh‑Go handle large model files?
In Local Mode, model files are streamed from the API on demand, so the phone never stores the full model. Remote Mode can offload large models to a server with GPU acceleration.
Is the SSH connection encrypted?
Yes, dsh‑Go uses standard SSH with key‑based authentication, providing end‑to‑end encryption for all remote commands.
Can I integrate custom plugins with dsh‑Go?
Absolutely. Place plugin scripts in the workspace directory; the harness will load them at startup just like on a desktop installation.
Neptune Infotech can help you integrate dsh‑Go into your enterprise mobile strategy, ensuring robust AI capabilities and secure deployment.