Why it matters
This PWA provides AI builders with on-the-go access to their DeepSeek Harness instances, enabling them to manage sessions, interact via streaming chat, and handle approvals directly from their mobile devices. The zero-install approach lowers the barrier to entry for mobile interaction with AI agents.

What changed

A new Progressive Web App (PWA), named dsh-pwa, has been developed to provide mobile access to the DeepSeek Harness (dsh) platform. This PWA mounts at the /m path on a dsh web server and offers functionalities such as session listing, real-time streaming chat, approvals, agent questions, new session creation, and image messaging. A key design choice is its PWA nature, which eliminates the need for users to install a native application on their iOS or Android devices; instead, users can open it in mobile Safari or Chrome and add it to their home screen for a full-screen app-like experience.

The dsh-pwa is positioned as a zero-install alternative to native mobile applications, specifically targeting users who need functionality on both iOS and Android without app installation. It contrasts with the saya-ch/dsh-mobile native Android app, which offers system notifications and certificate-pinned device pairing, and the official web UI which is desktop-first. The PWA inherits its remote access capabilities from the dsh web server, supporting LAN, Tailscale, or any reverse proxy configuration. However, it notes limitations such as the absence of push notifications on iOS PWAs due to platform constraints.

Key features of the dsh-pwa include a session list with workspace grouping, search, pull-to-refresh, and an "needs attention" filter. The streaming chat supports optimistic sending, queued message visibility, per-day dividers, timestamps, code-copy buttons, and a full-screen image viewer. Approvals and agent questions are presented as tappable cards, with dangerous commands requiring a double confirmation. Users can send images from their camera, photo library, or clipboard, with automatic compression to model limits. New sessions can be created with an agent-preset picker. Additional features include haptics, dark/light themes, per-session draft autosave, and a connection status indicator with manual reconnect. The desktop GUI also receives an update with a "手机端" (mobile) entry in its settings, displaying the phone URL for easy access.

Installation involves adding dsh-pwa as a dependency to a dsh web profile's package.json, running npm install, and then starting the dsh web server with appropriate --trusted-host configurations for remote access. The PWA works by serving static files from the web/ directory at /m and communicates directly with DSH's /api endpoint, using the same protocol as the desktop GUI. This design minimizes business logic duplication and ensures the PWA moves with DSH version updates. The project is developed in Vanilla JS with no build step or framework, aiming for simplicity and ease of contribution.

Why it matters for builders

This PWA offers AI builders a convenient way to interact with their DeepSeek Harness instances directly from their mobile phones, without the overhead of installing native applications. It enables continuous management of AI agent sessions, real-time chat interactions, and approval workflows, even when away from a desktop environment. The direct API communication ensures that the mobile interface remains synchronized with the core DSH functionality.

Practical impact

AI builders can now leverage their mobile devices for immediate oversight and control of their DeepSeek Harness deployments. This includes responding to agent queries, approving critical actions, and monitoring ongoing sessions from anywhere with an internet connection. The quick setup via npm and browser access means builders can start using the PWA with minimal friction. Developers interested in contributing can engage with the project's Vanilla JS codebase and contribute to roadmap items like session management or web push notifications.

Caveats and source limits

The dsh-pwa currently lacks push notification support on iOS due to PWA platform limitations. While it aims for seamless integration, the project is still in a developer preview phase, and its API may undergo breaking changes, requiring corresponding updates to the PWA. Offline functionality is not supported, as it requires a live connection to the dsh server. The security model relies on the dsh web server's existing trusted-host fence, meaning exposure should be limited to trusted networks to prevent unauthorized access to the agent. The project is open-source, with a MIT license, and contributions are welcomed via pull requests.

Sources

Written with AI assistance from the linked sources; every claim below was checked against them automatically. How we produce articles.

Claim check: 7/7 supported claims - 7 evidence links - 100% avg confidence
  • dsh-pwa is a phone-first PWA for DeepSeek Harness, offering session list, streaming chat, approvals, agent questions, new sessions, and image messages.supported - github.com
  • The dsh-pwa requires no app installation and works on iOS and Android via mobile browsers by adding it to the home screen.supported - github.com
  • The PWA communicates directly with DSH's /api endpoint, using the same protocol as the desktop GUI.supported - github.com
  • The dsh-pwa does not support push notifications on iOS due to platform limitations.supported - github.com
  • The dsh-pwa does not work offline and requires a live connection to the dsh server.supported - github.com
  • The dsh-pwa is developed in Vanilla JS with no build step or framework.supported - github.com
  • The dsh-pwa is licensed under the MIT license.supported - github.com

Caveats

  • Single-source caution: verify critical details at the linked source.
Radar score 86/100 - how it was calculated
Reliability82
Freshness92
Novelty77
Technical81
Developer96
Ecosystem72
Confidence100
  • Reliability 82: GitHub metadata supports source trust
  • Freshness 92: Fresh GitHub activity
  • Novelty 77: Fresh GitHub release
  • Technical 81: Repository technical metadata
  • Developer 96: Developer tooling signals
  • Ecosystem 72: Fresh GitHub release
  • Confidence 100: Claims have reliable evidence
Share
XLinkedInHacker News

Related articles

AI Tools - Sep 29, 2026DSH Skill Trace: Agent Skill Loading Visibility PluginPolinniZhong has released DSH Skill Trace, a local-first plugin for DeepSeek Harness that enhances visibility into which skills an agent loads during a conversation. The plugin provides detailed receipts and flow maps of skill invocations, allowing users to understand, review, and learn from the agent's operational process.AI Tools - Sep 29, 2026AgenticOS: Open-Source Platform for Building and Managing AI AgentsAgenticOS is a new open-source, self-hosted platform designed for building, running, and governing AI agents within an organization. It provides a unified environment for managing agent skills, context files, automations, and budgets, with a focus on auditability and control.AI Tools - Sep 29, 2026OpenSider for VS Code Integrates Multiple AI AgentsOpenSider for VS Code is a new open-source extension that allows developers to use multiple AI agent CLIs, such as Claude Code, Codex, Cursor, OpenCode, and GitHub Copilot CLI, from a single VS Code side panel. It offers a unified interface for these agents without bundling any models itself.AI Tools - Sep 29, 2026TruePlumb: Open-Source Testbed for AI Agent Safety ControlsTruePlumb is an open-source, vendor-neutral testbed designed to verify the effectiveness of AI agent safety controls. It uses deterministic, judge-free verification methods based on linear temporal logic to assess guardrails, firewalls, and other security mechanisms.AI Tools - Sep 29, 2026Claude Command Center v5.35.0 ReleasedClaude Command Center has released version 5.35.0, a Python-based open-source tool for managing AI agent sessions locally. The tool supports orchestrating sessions from various AI models and services.AI Tools - Sep 29, 2026HProxy Free Proxy List Integrates with AI AssistantsThe HProxy free proxy list project now offers direct integration for AI assistants and agents, providing a keyless API and specialized tools. This allows AI applications to easily access and utilize a continuously updated list of free proxies with filtering capabilities.