Flaq AI Agent Guide

Practical Guide for Using DeepSeek Harness with Flaq AI's OpenAI-Compatible API

Run the open-source DeepSeek Harness Web UI with Flaq AI's OpenAI-compatible endpoint, then connect a supported DeepSeek v4 model for repository analysis and coding workflows. DeepSeek Harness follows an “everything is a plugin” architecture: model adapters, tools, the agent loop, session storage, permissions, sandboxing, telemetry, and the Web UI can be composed or replaced through plugins and configuration.

Agent guides

Claude Code/cc-apiOpenAI Codex/codex-api/v1Hermes Agent/api/v1DeepSeek Harness/api/v1ZCode/api/v1

Install

Run DeepSeek Harness

DeepSeek Harness requires Node.js 22.19+ or a 24+ release. Start it from the project directory you want to use as the default filesystem location; the official npm command launches the Web UI at http://127.0.0.1:3080 by default.

1

Run the DeepSeek Harness Web UI

Use the official npm package to start the Web UI. Keep this terminal running while you use DeepSeek Harness.

The first run downloads the current @deepseek-ai/dsh package and initializes the Web UI profile.

Official npm command

npx @deepseek-ai/dsh web

Repository: Review the latest installation instructions, releases, compatibility notes, and source code in the official DeepSeek Harness GitHub repository.

Configure

Connect DeepSeek Harness to Flaq AI

Open Settings → Models in DeepSeek Harness, add Flaq AI as a custom provider, and enter your Flaq API key. The Provider ID is permanent after saving, so use flaq unless you intentionally need a different route name. After saving the provider, choose a workspace before starting a session.

Recommended DeepSeek v4 Pro model for repository analysis, coding tasks, and workflows that benefit from stronger reasoning.

In Settings → Models, select Add provider → Add a custom provider. Enter the values below and add the selected model ID manually instead of using Fetch available models. Save the provider, then select that model for a new session. DeepSeek Harness treats manually entered models as text-only unless additional modalities are declared.

  • Use the exact Flaq base URL and openai-completions protocol.
  • Replace the placeholder with your Flaq API key, and keep the model ID exactly as shown.

Flaq AI custom provider

Provider ID: flaqDisplay name: Flaq AIBase URL: https://api.flaq.ai/api/v1API protocol: openai-completionsCredential / API key: Your_Flaq_Api_KeyModel ID: deepseek-v4-pro-text-to-text

Verify

Verify Flaq in DeepSeek Harness

Open the Web UI, choose the project directory where you started DeepSeek Harness, select the configured Flaq model, and create a new session. The composer stays unavailable until a workspace is selected.

  • DeepSeek Harness prints the active URL in the terminal.
  • The default local address is shown below.

Open the Web UI

http://127.0.0.1:3080
  • Send a read-only task first.
  • A normal answer that reflects the selected workspace confirms that DeepSeek Harness can reach the Flaq endpoint and use the configured model.

Test repository access

Summarize this repository and identify its main packages. Do not modify any files.

Plugin Example

Install a DeepSeek Harness Plugin

Use DSH OpenPencil as a concrete example of DeepSeek Harness's “everything is a plugin” architecture. Complete the Flaq provider configuration and verification above, stop the running Web UI, then add the plugin to the web profile and restart the same DeepSeek Harness version. DSH OpenPencil adds tools for creating, previewing, inspecting, and editing .op design documents inside a conversation.

1

Install DSH OpenPencil

Add the public OpenPencil package to the web profile with the DeepSeek Harness version currently tested by the plugin. Run this command from the same project directory you use for DeepSeek Harness.

Install DSH OpenPencil

npx --yes -p @deepseek-ai/[email protected] dsh plugin --profile web add @zseven-w/dsh-openpencil@latest

Repository: Review the source code, setup requirements, releases, and compatibility notes in the DSH OpenPencil GitHub repository.

2

Restart the Web UI

Start the tested DeepSeek Harness version again so the web profile loads DSH OpenPencil. Keep this terminal running while you reopen the Web UI.

Restart the Web UI

npx --yes -p @deepseek-ai/[email protected] dsh web
3

Test DSH OpenPencil with Flaq AI

Select the configured Flaq model and workspace, start a new session, and send the request below. When the OpenPencil binary is available, the plugin should create an .op document and present an editable preview in the conversation.

Test DSH OpenPencil with Flaq AI

Create a simple editable OpenPencil document with a title, a subtitle, and two feature cards. Save it as openpencil-demo.op, then render an editable preview.

Open Source

Explore the Official DeepSeek Harness Repository

DeepSeek Harness is an open-source agent harness developed by DeepSeek AI. Visit the official repository to review the source code, follow releases and compatibility changes, read the latest documentation, and report issues during the developer preview.

DeepSeek Harness on GitHub

Open the official DeepSeek Harness repository for source code, setup documentation, development notes, discussions, and project updates.

Open-Source Guide

Explore Flaq AI's Open-Source DeepSeek Harness Guide

Go beyond the initial setup with Flaq AI's independent, community-maintained guide to DeepSeek Harness. Learn how its plugin tree, profiles, bundles, patch layers, replaceable agent loop, model adapters, tools, session services, permission policies, sandboxes, and Web UI extensions support custom agent workflows.

DeepSeek Harness Guide on GitHub

Explore practical architecture notes, a usage handbook, plugin development recipes, multilingual documentation, and reusable agent skills for understanding, extending, and building on DeepSeek Harness.

Troubleshooting

DeepSeek Harness FAQ

Using this guide's setup may significantly increase token usage.

DeepSeek Harness Discussions on X (Twitter)

Explore discussions by developers and early participants about DeepSeek Harness (DSH), an open-source agent harness, covering its “Everything is a Plugin” architecture, composable runtime, plugin ecosystem, and hands-on experiences.