Systems / agent workWORK / AGENTIC-OS

Agentic OS

A connected workspace for agent work.

A local-first workspace that connects an agent’s task to the context, skills, tools, and evidence behind its work.

Agentic OS concept interface: a radial workspace with fictional projects and resources, with an active thread connected simultaneously to context, skills, and MCP tools, plus linked evidence and an outcome.
01 / A connected workspaceConcept interface with fictional content, created for this overview.

The problem

Agent work spreads across projects, skills, context files, and servers. Without one place to hold them, every run starts by rebuilding the setup by hand.

The approach

Keep the active thread connected to its project, context, skills, and MCP tools at the same time. Evidence and artifacts stay attached to the work. A radial graph makes those relationships visible in one workspace.

02 / Follow the work

One thread. Many connections.

Context, skills, and tools are connected at the same time. Scroll to explore each relationship while the whole workspace stays in view.

Agentic OSWorkspace mapConcept view
Radial Agentic OS workspace. The active thread is connected simultaneously to its project, context, skills, and MCP tools, with evidence and an outcome attached. Reading focus: Active thread.AGENTIC OSContextProject briefSkillsWriting guidelinesMCPFind referencesEvidenceSources attachedOutcomeField guideActive threadPrepare a field guideONE WORKSPACE / ALL RELATIONSHIPS IN VIEW

Scroll to continue Or choose a connection above

01 / Active thread

One thread, connected to everything it needs.

The active thread is a working session inside a project. Its context, skills, and MCP connections are available together. The graph shows those relationships around the thread while the agent works.

In this examplePrepare a field guide for a new team member.

02 / Context

Context gives the thread its bearings.

Context is the knowledge connected to the thread: the project brief, reference notes, and decisions that make the request specific. It sits alongside skills and tools throughout the work.

In this exampleA project brief, a glossary, and a few useful notes.

03 / Skills

Skills shape how the agent works.

Skills are reusable instructions associated with the thread. They can describe writing conventions, a way to approach a problem, or the checks a result needs. The agent uses them together with its context and tools.

In this exampleWrite plainly. Organize for a first read. Check every reference.

04 / MCP

MCP puts tools within reach.

MCP connects the active thread to tool capabilities. Those connections are present alongside context and skills; the agent can use a tool when the work calls for it. Availability and an actual tool call are different things.

In this exampleFind a reference. Read the selected source. Prepare a draft.

05 / Evidence

Evidence stays attached to the thread.

As work happens, sources, tool results, and review notes form an evidence trail. They belong to the thread that produced them, making it possible to inspect what informed a result and what was checked.

In this exampleReferenced notes, a draft artifact, and a completed review.

06 / Outcome

An outcome with its working context intact.

A deliverable belongs to the same connected workspace. Someone reviewing the field guide can also inspect its evidence and the resources associated with the thread. The graph preserves the relationships around the result.

In this exampleA field guide, ready for a person to review.

The idea behind the graph

The result matters.
So does what supports it.

A connected record makes agent work easier to understand, review, and continue. The graph brings the request, the resources, and the evidence into one picture.

More tools, systems, and experiments.

Back to all work

Loading

Still loading. The connection may be slow. This page could not load. Check the connection and reload.