FAQ
Updated on 2026-10-15 · Estimated reading time: 5 minutes
Usage
How are credits consumed?
Credits are consumed in the following order:
-
The batch that expires earliest is deducted first.
-
If batches expire on the same date, bonus credits are deducted first, then plan base credits, then add-on package credits.
-
For successful, failed, or interrupted tasks, actual resource consumption is settled first.
See Credit Rules for details.
Why can’t AI find a file in the workspace?
Check the following:
-
Whether the conversation uses the correct workspace.
-
Whether the file is in the workspace root or an authorized folder.
-
Whether workspace permission has been removed.
-
Whether the file still exists locally and has not been moved.
-
Whether the current agent can read files.
Why is an expert, Skill, or model not visible?
Possible reasons include:
-
You are in a different account space.
-
The enterprise administrator has not configured or authorized it.
-
The Skill is not installed.
-
MCP or an extension is not connected.
-
The resource is disabled or unsupported in the current version.
Why does a task take a long time?
A complex task may need to read multiple files, call tools, and perform multi-step reasoning. Check whether deep reasoning is enabled, whether the context is too large, whether a tool is waiting for an external service, and whether a step requires human confirmation.
How do I avoid generating files in the wrong project?
Confirm the workspace before starting the task. Specify the output directory and file name in the prompt, then check the file location after generation. Use separate workspaces and conversations for different projects.
What should I do when AI output contradicts the source material?
Stop using the result directly. Ask AI to list its sources and reasoning, compare the result with the source material, and if necessary narrow the task and generate it again. Important conclusions should be confirmed by the accountable business owner.
Does DeepWorks run locally or in the cloud?
DeepWorks uses a local-first, local-cloud collaborative product form. The desktop app carries files and task context in local workspaces, while model services, Cloud Workers, and enterprise systems can be called according to enterprise configuration. The actual execution location and data path depend on deployment and enterprise configuration.
Are local files and business data uploaded to the cloud?
DeepWorks accesses local files only within the scope authorized by the user. Whether task content is sent to a cloud model or another service depends on the selected model, extensions, MCP, and deployment. For sensitive data, confirm enterprise data-security policy and the relevant service configuration first.
What is the difference between the general assistant, experts, and expert teams?
The general local AI assistant is suitable for daily questions and general tasks. An expert focuses on a single professional domain. An expert team is suitable for complex tasks that require multiple roles to collaborate.
What is the difference between Skills, MCP, and plugins?
Skills capture standardized, reusable working methods. MCP connects external tools, enterprise data, or business services. Plugins add capabilities such as browser control, computer control, and image generation to DeepWorks. All three may be called by agents or tasks, but their installation, connection, and authorization methods differ.
Why do different accounts see different models, experts, and Skills?
Different accounts may belong to different personal or enterprise spaces and have different roles and permissions. Models, experts, Skills, MCP, and enterprise knowledge may also be configured by administrators for specific spaces or people, so different visibility is expected.
Which operations normally require human confirmation?
Operations that write or modify business-system data, delete or overwrite files, send external emails or messages, control a browser or local computer, force-push code, or call external services involving sensitive data should normally require confirmation. The exact policy follows enterprise security configuration.
System Issues
Windows shows “Smart App Control blocked an unsafe app”
This block comes from the Windows security mechanism. Add an exclusion in Windows Security: Windows Security → Virus & threat protection → Ransomware protection → Allow an app through Controlled folder access, then reopen DeepWorks.
macOS shows the app is damaged and cannot be opened
This is caused by macOS Gatekeeper. Remove the quarantine attribute in Terminal, then reopen the app.
xattr -cr /Applications/DeepWorks.appThe Linux .deb installation does not open
If the app does not respond after installing the .deb package, or starting it from a terminal reports a sandbox error, the chrome-sandbox permissions are usually incorrect.
Run:
sudo chown root:root /opt/DeepWorks/chrome-sandbox
sudo chmod 4755 /opt/DeepWorks/chrome-sandboxThen reopen DeepWorks from the application menu, or verify with:
/opt/DeepWorks/deepworks