LoneAct Documentation LoneAct Documentation

Core concepts

LoneAct organizes long-running work around a few stable ideas: project folder, conversation, run, Agent team, and workbench.

LoneAct in one picture

LoneAct keeps project context and execution evidence in one workbench. You do not need to remember where every page lives; just follow the path from project → conversation → run → review.

Project folder

A project folder you authorise, opened by the LoneAct service on this computer. It defines the project identity, file scope, and where tools may run.

Where work starts

Conversation

An ongoing conversation and execution record that accumulates around one goal. It can be split, restored, and continued.

How work is recorded

Run

One Agent execution. It can include model generation, tool calls, approvals, sub-agents, and a final review.

How work is executed

Agent Team

The network of roles and digital employees that belongs to a project. Roles define responsibilities, and Agent instances keep their own long-term memory.

Who does the work

A device can have more than one project

One device can own multiple project folders, and each folder is an independent Project Location. With remote access enabled, local projects and projects on authorised devices can appear in the same Workbench; folders with the same name are never merged automatically.

Multiple projects

Browse projects by device in Locations; every project keeps its own Conversations, Agent Team, Memory, Rules, and checkpoints.

Local and remote

Local Web and the IDE can connect to a local Direct Endpoint and a remote Cloud Relay at the same time; tools always run on the device that owns the project.

See Projects and devices for the full model, grant boundaries, and a mixed-workbench example.

The Workbench is the only work surface

Conversations, project overview, Agent Team, tools, settings, and checkpoints are not separate products; they are different views of the same Workbench. Switching entry points does not change the project identity or move execution records somewhere else.

Surface you seeQuestion it answersCommon actions
ProjectsWhich projects am I working in?Open a Location, check status that needs attention, continue a conversation
ConversationWhere is this goal now?Send a goal, approve tools, review diffs and tests
Agent TeamWho owns which capabilities?Review roles, edit responsibilities, update the team from code
SettingsWhat are the models, keys, and boundaries?Switch the default model, configure Secrets, manage remote access

What local-first means

In local or private mode, project files, conversations, run records, Agent memory, and execution logs are stored on your computer. The cloud account service only handles accounts, devices, access grants, and necessary audit metadata; it is not a business data warehouse.

Business data stays local

Unless you explicitly choose cloud execution or call an external model, LoneAct does not upload complete project files to the cloud control plane. Model requests are still sent to the provider you select; see Data and local-first for the exact boundary.

Roles, Agents, and digital employees

A Role is a job definition: it describes responsibilities, instructions, available models, skills, and delegation depth. An Agent is the person actually doing that job, with its own directory, long-term memory, and identity across conversations. Combine the two and you get a digital employee that can collaborate over time.

Role

A reusable responsibility template for stable, auditable team structure.

Agent

A long-term collaborator with memory, suited to owning one domain continuously.

Team

A network of roles built around project capabilities that can be generated, revised by hand, or rebuilt.