StaffDeck Update | From a Single Digital Employee to a Collaborative, Executable Enterprise AI Team
On July 15, StaffDeck 0.1.0 was officially open-sourced, and in the month that followed, the project completed more than 270 commits. Today, version 0.4.0 is officially released!
The first version solved the creation and management of digital employees, enabling professional knowledge, business experience, and SOPs to be used continuously by digital employees. From 0.2.0 to 0.4.0, three releases continued to move toward real enterprise work scenarios — digital employees began to form teams, connect to common chat tools, call external tools, and integrate with enterprise business systems.
Let's take a look at the key changes in StaffDeck compared with 0.1.0.
From individual employees to team collaboration
This update introduces multi-agent group chat, allowing users to organize individual digital employees into teams on top of creating and managing them.
Team collaboration now works as follows:
When creating a team, you can set a project lead and team members.
When a complex task enters the team, the project lead breaks it down, then selects executors through direct assignment or bidding.
Once members are selected, they can handle their subtasks asynchronously without blocking each other.
The team group chat shows replies and collaboration progress, and key materials and conclusions can be shared on the team whiteboard.
When work is done, the project lead is responsible for acceptance. If the result does not meet requirements, it can be sent back for adjustment, and users can take over at any time when human judgment is needed.
Each task has its own execution thread, so members don't crowd into the same ever-growing context. The system records how each task was broken down, who ultimately executed it, and how acceptance went, making it easy to review and trace later.
Task execution with downloadable deliverables
This round of updates rolls out the Harness v2 execution kernel across the board. Conversations, SOPs, scheduled tasks, and the open API now reuse the same execution capabilities.
After receiving a task, a digital employee can dynamically discover the knowledge, skills, and tools it needs, read materials and process data in an independent TaskFrame workspace, run controlled commands when necessary, and finally generate business files.
To ensure secure operation in enterprise environments, StaffDeck has added HTTP API, MCP, and MCP Apps tool integration, as well as file read/write and directory operations. Each task's workspace is isolated, and execution duration, output size, and network access all have explicit limits; on macOS, Linux, and other environments, system-level sandboxes can further tighten the scope of task activity.
Reports, spreadsheets, Markdown, and other files generated after task execution can be downloaded by users after security validation and registration.
Unified access across WeChat, WeCom, Feishu, and DingTalk
Many needs in an enterprise start with a chat message. If digital employees can only live on a standalone web page, every use takes an extra step.
StaffDeck has now built a unified channel integration kernel and supports WeChat, WeCom, Feishu, and DingTalk. Employees can initiate requests directly in the chat tools they already use, while the same digital employee capabilities and Harness v2 handle them behind the scenes.
A channel account can connect to multiple digital employees. The system selects the appropriate employee based on message content, and users can view the current employee and switch manually. Channel identities can be merged with StaffDeck accounts. When an employee moves from the web to IM, the original conversation can continue, and related memory and execution records stay connected.
Channel services also add basic capabilities such as message deduplication, retry on failed delivery, and recovery after service restarts. Administrators can view delivery logs and running status. Channels such as Feishu and DingTalk now support receiving attachments, and some channels use emoji to indicate that a task is being processed.
Open API brings digital employees into business systems
Beyond the chat entry point for employees, StaffDeck also provides a complete open API so business systems can call digital employees directly.
Enterprises can create an API Key for the entire account or a separate key for a specific digital employee. CRM, ERP, ticketing systems, and internal platforms can all initiate tasks through the API, placing digital employees into existing business processes.
The open API now covers the main stages of a digital employee's lifecycle, from management to execution.
Enterprises can create and manage digital employees through the API.
Conversations support both stateful and stateless modes.
Runs can execute synchronously in streaming mode or be converted to asynchronous tasks. Callers can receive progress and replies via SSE and continue reading after a disconnect.
Tasks can be canceled during execution, and results can be queried and artifacts downloaded after completion.
SOPs, knowledge, skills, tools, and scheduled tasks can all be managed through the API, and Webhooks send key events back to business systems.
Idempotent requests and fine-grained permission control help business systems handle retries and concurrent calls.
On permissions, employee-level keys can only call the specified digital employee and cannot read model keys or plaintext tool credentials. An API Key's plaintext is shown only once at creation or rotation.
Knowledge and SOPs enter a phase of continuous evolution
0.1.0 already provided structured SOPs and a document knowledge base. The current version continues to solve two problems: more accurate retrieval and updatable knowledge.
Changes to the knowledge module focus mainly on retrieval and maintenance:
Document parsing preserves the original section and hierarchy structure as much as possible.
Queries can narrow the knowledge scope and control the amount of relevant evidence, reducing irrelevant content in answers.
Retrieval across multi-turn conversations stays continuous.
Citations and sources are managed uniformly, making answer grounding easier to verify.
Users can view relationships between content through the knowledge graph and directly edit accumulated knowledge.
SOPs now have an independent workspace. Users can edit and debug flows there, and publish to the plaza once they confirm normal execution. SOPs already in use can still be modified, and updated versions can be reused by other digital employees.
Cross-platform delivery and stability
To help more users actually get StaffDeck running, the current version further improves cross-platform delivery and runtime stability.
The current version supports the following deployment options:
macOS Apple Silicon client;
macOS Intel client;
Windows x64 installer;
Linux x86_64 Debian/Ubuntu package;
source deployment on macOS, Linux, WSL, and Windows PowerShell.
Desktop installers gradually bundle fixed versions of Python, Node.js, and the sandbox runtime, reducing local environment differences. The Linux version supports local, LAN, and public network modes, and the desktop client prompts for upgrades when a new version is found.
On model integration, StaffDeck adds Anthropic, Gemini, and other protocols alongside the OpenAI-compatible protocol, and improves custom model addresses, Basic Auth, custom request headers, and MCP credential configuration.
Meanwhile, we have continued to improve:
conversation task cancellation and queuing;
model configuration validation and atomic saving;
session execution records and latency observability;
batch JSON log export;
channel crash recovery and delivery retries;
user avatars, account menu, and quick-start experience;
packaging reliability on Windows, macOS, and Linux;
security for external links, credential storage, and artifact access.
0.1.0 proved that professional knowledge, business processes, and tool capabilities can be consolidated into digital employees.
The current version goes further: digital employees begin to form teams, enter enterprise IM, call external tools, and connect to business systems through APIs. StaffDeck is evolving from a digital employee building tool into an intelligent collaboration and execution platform for enterprise business processes.
Our goal is not just to make AI "better at chatting," but to let organizations truly build, run, and govern their own digital employee teams.
How to get it
Visit the StaffDeck official website:
🔗 https://staffdeck.openbmb.cn/
Go to GitHub for the code and latest Release:
🔗 https://github.com/OpenBMB/StaffDeck
We welcome you to try it, Star it, and share your use cases in the community.