Product and capabilities
Is LoneAct only a coding tool?
No. It starts from code work, but the product model is a general Agent workbench: project folders, Agent teams, tools, memory, and permissions can carry long-running data analysis, operations, support, and research work.
How is it different from ordinary AI chat?
Ordinary chat mainly produces text; LoneAct lets Agents call tools, modify files, run tests, and leave execution evidence inside project boundaries. A conversation is a work record, not just a list of messages.
Does it support multiple models?
Yes. The workspace has a default model, conversations can switch models, roles can specify different models, and supported models can use your own model key.
Is the Agent Team fixed?
No. Planner / Executor / Reviewer is only a common starting point; teams can be built from project capabilities, extended by hand, revised, or turned into digital employees with long-term memory.
Security and data
Are secrets sent to the model?
No. Provider keys and project Secrets are held in LoneAct secure storage; the model only sees variable names or secret:// references, and the actual value is resolved when a tool executes.
Do local environment variables count as secrets?
Yes. Environment variables configured through Secrets do not enter chats, prompts, or model context; they are available in the process environment the Agent starts but are not read by the model.
Where is business data?
In local or private mode, project files, conversations, run records, Agent memory, checkpoints, and execution logs are stored on your computer; the cloud account service stores only accounts, devices, access grants, and necessary audit metadata.
Does the cloud keep full conversations?
Cloud Chat is for temporary runs and does not write full chat history as cloud business data. Goals and execution records worth keeping belong in a project conversation in the local Workbench.
Product surfaces and usage
Which entry point should I start with?
Start with Local Web for a first run. Use the IDE when you want to stay close to code, the browser extension when you need page context, and the cloud workbench when you need to continue remotely.
Does data diverge between entry points?
No. Project identity and conversations are determined by the project folder and the device that owns it; an entry point is only an interface and does not turn one project into two sets of business data.
Can the cloud touch my local files?
Only when the device that owns the project is online and remote access has been granted; the cloud itself cannot read your local files directly.
How do I start building digital employees?
First understand the difference between Role and Agent in Settings → Agents, create an Agent with memory, then add it to a project Team. See Agent teams and digital employees.
When something goes wrong
- Model unavailable: check the plan, model catalog, and provider connection, then switch to an available model.
- Project not visible: confirm your device is online, the project is authorized, and the current account is correct.
- Tool denied: review the approval prompt and execution boundary, narrow the goal, and try again.
- Remote failure: confirm your device is online, the access grant has not expired, and project permissions cover the current action.
- Secret-related failure: confirm the Secret is enabled, the variable name matches, and restart affected processes after a change.
Continue with the matching chapter in the docs navigation, or check your device, model, and tool status in the product. Never send real secrets or production credentials through chat.