Skip to content
Self-Hosted AI Agents with BISHENG: Open-Source Coze Alternative for Enterprises

Self-Hosted AI Agents with BISHENG: Open-Source Coze Alternative for Enterprises

The author deploys the open-source BISHENG v3.0.0 platform via Docker Compose, connects its desktop client, builds workflows, and demonstrates enterprise features like model permissions, usage tracking, and Vibe app hosting as a self-hosted alternative to Coze.

Sohu Tech Products
Sohu Tech Products
Sohu Tech Products

Coze announced on October 29 a product integration upgrade that sets a deprecation timeline for existing enterprise plans. The author notes that Coze was widely adopted — for example, a neighboring company in a second-tier city used it for enterprise deployment — because it combined knowledge bases, plugins, and workflows in a single drag-and-drop interface, making it many users' first exposure to AI agents.

This shutdown prompts a reevaluation: enterprises need long-term control over their knowledge, processes, and deployment environments. The author searched for open-source alternatives and found BISHENG (GitHub: https://github.com/dataelement/bisheng), which has 12k stars and active issues, PRs, and contributors. BISHENG is an open-source, privately deployable enterprise-grade Agent platform that also functions as an AgentOps system focused on office productivity. It lets organizations host models, knowledge, tools, and permissions in their own environment. The same team previously built DSH Desktop, a third-party DeepSeek client that grew from 230k to 930k+ downloads in two months; the BISHENG desktop client is built on top of DSH Desktop.

Installation and Deployment

The author deploys BISHENG v3.0.0 on macOS using OrbStack (Docker Compose). The manual steps are:

git clone --branch v3.0.0 --depth 1 https://github.com/dataelement/bisheng.git
cd bisheng
bash docker/install.sh

The script prepares configuration, pulls images, and starts services. Alternatively, the author provides a prompt for an AI assistant to automate the deployment, including checking system architecture, memory, port usage, selecting compatible images, preserving existing Docker services, and verifying health until the registration page loads at http://127.0.0.1:3001/. The first registered user becomes the administrator.

Enterprise DSH Core and Desktop Client

The web UI is the enterprise platform; the desktop client connects to it. The relationship: DSH Desktop is the base desktop agent, while the BISHENG client adds enterprise platform connectivity. The web platform handles building and management; the desktop client lets employees consume platform resources for daily work. The author configures a DeepSeek API key in the desktop client's model settings. Because OrbStack provides an HTTPS address for the BISHENG container, that address is entered as the enterprise platform URL in the desktop client to enable login. A sample AI prompt for this connection is shown.

Chat/Work Dual Modes

The desktop client offers two modes:

Chat mode : simple Q&A, fast response, data synced across web, desktop, and mobile.

Work mode : complex tasks using local files continuously; includes Office-suite template capabilities. The author demonstrates Work mode by generating a PPT from 2024-2025 Li Auto/Xpeng/NIO sales data using the Blue Professional template, noting the visual quality exceeded their company's annual reports.

Building a Workflow

Back in the enterprise web UI, the author explores the build (workflow) section. BISHENG includes application templates; the author selects an "Article Assistant" template, which reveals many customizable nodes. They then create a custom workflow using tool nodes and run it to analyze the DeepSeek-V4.1-Flash paper.

Hosting Vibe Applications

The platform can one-click host "Vibe" applications (apps generated by coding agents) for employee access and unified management. The author argues that while desktop agents handle personal tasks, teams continuously generate collaborative needs (e.g., training sign-up forms, project ledgers, department dashboards) that are better served by dedicated apps. Developers can use familiar tools or coding agents locally, then submit to BISHENG where the app passes pre-check, secret scanning, and build/release approval before becoming a shared business tool. A screenshot shows non-engineering team apps already running on the platform.

Unified Model, Usage, and Permission Management

The enterprise UI provides:

Model permissions : assign models by department and role.

Usage tracking : view consumption by department and role.

Quota allocation : assign credits (e.g., 10k credits to self) with limits.

Audit and approval : automatically review employee-sent content, AI responses, and tool-call parameters for sensitive or classified data.

Organization structure : manage hierarchy.

Knowledge bases, datasets, and model fine-tuning are also supported.

Conclusion

After installing, connecting the desktop client, creating a PPT, and running a workflow, the author summarizes BISHENG's value: enable employees to use AI in daily work while giving enterprises centralized management . Employees operate on files via the desktop client; the company controls models, knowledge, tools, and permissions. Individuals can use the open-source DSH Desktop personal edition; companies deploy the BISHENG platform and configure it to their needs. The author emphasizes that while tools change, accumulated knowledge and business processes should remain under the organization's control, and notes the BISHENG team is actively maintaining the project.

Related links:

GitHub: https://github.com/dataelement/bisheng Official site: https://www.bisheng.ai/ DSH Desktop:

https://dshdesktop.com/
Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

WorkflowOpen SourceAI AgentCozeEnterprise AIdocker-composeSelf-HostedBISHENG
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.