MIT-licensed, self-hosted browser sandbox (Python / FastAPI, React, Docker) for AI agents: CDP automation, noVNC / Xpra live view, persistent profiles and human takeover.
DSH integration
Compatible
Author-claimed
Safety audit
Unaudited
Last verified
2026-11-10
License
MIT
01What can it help you accomplish?
Give a DeepSeek Harness agent a real Chromium browser on a Linux server instead of on your own machine
A Docker-hosted GUI Chromium sandbox that the agent drives over CDP (Playwright / Puppeteer, SDK, CLI or REST API)
Teams running browser-using agents that need an always-on, isolated browser
Watch the agent browse and step in when it reaches a login screen or gets stuck
A live noVNC / Xpra view where automation is paused before you take over
Operators who have to intervene during agent runs
Keep browser state between runs instead of starting from a blank profile
A per-sandbox browser profile and shared files preserved across pause and resume
Agent workflows that depend on logged-in sessions or previously downloaded files
02How to install into DeepSeek Harness
Prerequisites
- A Linux Docker host with Docker Compose; the host does not need a desktop environment
- The admin token configured for the service, needed to sign in at /admin
Installation steps
- 01
Set up the service using the server deployment guide (docs/guides/headless-linux.md), or follow the local Docker setup (docs/guides/deployment.md)
- 02
Open /admin at your service address and sign in with the configured admin token
- 03
Create a sandbox, open its visual session, or connect your agent using the usage guide (docs/guides/usage.md)
Verify the integration
Not specified by the author
03DSH integration and capability boundaries
The repo ships a DeepSeek Harness preview plugin (packages/dsh-plugin) so a DSH agent can drive the sandbox; agents can also connect over CDP with Playwright or Puppeteer, or through the SDK, CLI and REST API.
Sandbox provisioning from the admin dashboard
the admin token and the /admin dashboard→sandboxes you can open visually or hand to an agent
Agent automation over CDP, SDK, CLI and REST
a sandbox's CDP endpoint (or the SDK / CLI / REST API)→programmatic control of the sandboxed Chromium
Live view and manual takeover
the sandbox's visual session→a noVNC or Xpra view of the running browser
automation has to be paused before you take overPersistent workspaces
an existing sandbox→its own browser profile and shared files, preserved across pause and resume
04Who is it for? When not to use it?
Good for
- Teams running browser-using agents that need an always-on, isolated browser
- Operators who have to intervene during agent runs
- Agent workflows that depend on logged-in sessions or previously downloaded files
Not for
- The current deployment model targets a single node in a trusted environment; the README points to the usage guide's compatibility and limits section for isolation and persistence boundaries.
05Compatibility, maintenance and safety notes
- The current deployment model targets a single node in a trusted environment; the README points to the usage guide's compatibility and limits section for isolation and persistence boundaries.
- Needs a Linux Docker host with Docker Compose; the host itself does not need a desktop environment.
- The original source is MIT, but the runtime images bundle separately licensed components, including Xpra under GPL v2 or later.
MIT · last push 2026-11-05; the facts pack records no GitHub release, so there is no published version tag to pin
06Frequently asked questions
How does Verge Browser connect to DeepSeek Harness?
The repository lists a DeepSeek Harness preview plugin (packages/dsh-plugin) under Integrations. Independently of that plugin, agents can connect through CDP with Playwright or Puppeteer, or use the SDK, CLI and REST API.
Do I need a desktop environment on the server?
No. The README requires a Linux Docker host with Docker Compose and states that the host does not need a desktop environment.
What do I do after deploying the service?
Open /admin at your service address, sign in with the configured admin token, then create a sandbox, open its visual session, or connect your agent using the usage guide.
Can I take over while the agent is browsing?
Yes. You can view the browser through noVNC or Xpra and pause automation before taking over.
How isolated is it, and what are the limits?
The current deployment model targets a single node in a trusted environment; the README points to the usage guide's compatibility and limits section for isolation and persistence boundaries.
07Related DSH workflows
archify
by tt-a1i
Turn any idea, plan, or codebase into a beautiful interactive diagram. An agent skill for Claude Code, Codex, and more.
openviking
by volcengine
Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.
aitoearn
by yikart
Let's use AI to Earn!
learn-harness-engineering
by walkinglabs
Harness engineering beginner tutorial, from 0 to 1
08Data and sources
| Integrations | [Codex](docs/integrations/codex.md), [Playwright MCP](docs/integrations/playwright-mcp.md), [DeepSeek H…
- **Agent automation**: connect through CDP with Playwright or Puppeteer, or use the SDK, CLI, and REST API.
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-11-10. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.