OpenLink / Agents

Keep every agent task in view.

Bring coding agents into your project, follow their tool activity, and review the changes in your own environment.

Self-hosted first · Codex / Pi

OpenLink AgentWorkflow illustration
atlas-web / project-list
Add a status filter to the project list. Keep the existing layout.

01Read the project

Locate the list, state and existing controls.

02Make a focused change

Connect the filter to the visible results.

03Check the outcome

Inspect the patch and running preview.

project-list.tsx

+ const [status, setStatus] = useState("all")

+ const visible = filterProjects(status)

Code changesBrowser resultYour review

Illustrative agent workflows

input
New Project Task

Start with a task and its project context.

Reading context...
Project brief
Repository files
Example
action
Scope & Plan

The agent reads the request and identifies the project files relevant to the task.

ExampleOpenLink Agent
output
Task Result

A focused plan is ready for you to inspect.

Answer

The task summary stays beside the project context, ready for the next step.

Plan outlined
Example

The agent engine is one part.
OpenLink connects the work.

OpenLink brings execution, project context and review surfaces together. Its Agent Host coordinates the environment, relays events and manages the session lifecycle around the underlying agent.

Project-bound execution

Bind the session to its project workspace and runtime. Environment preparation, worker launch and cleanup form one managed lifecycle.

Project
Session
Runtime

Visible tool activity

Follow execution events as the agent works. Read the actions and outputs without treating a status animation as proof of completion.

Request
Tool event
Result

Browser-connected work

Supported Pi setups bind browser tools to a session, bringing navigation, interaction and screenshots into the same task.

Agent
Browser
Review

Tools do the work.
You stay in the loop.

Follow commands and inspect results. In supported browser sessions, human input takes priority over Agent control.

A visible execution trail

openlink — bash · illustration

$ git status --short

M src/components/project-list.tsx

$ git diff --stat

src/components/project-list.tsx | illustrative patch

$ pnpm test

Review your project's actual test output before accepting a change.

Aceternity Terminal illustration. Not a live shell or a test-success claim.

Keep a hand on the controls

Explore the keyboard below. This is a tactile illustration, not a remote connection to your device.

Click a key, or focus the keyboard and type. Sound is off; key events stay inside this illustration.

Self-hosted first

Your environment. Your agent setup.

Cloud integration — coming soon

Choose where work runs

Self-hosted execution is the starting point. Available execution modes depend on your installed runtime and host configuration.

Choose an agent

Use Codex or Pi within OpenLink. Tools, extensions and browser capabilities vary by engine and setup.

Understand the data boundary

Self-hosting controls the execution environment. Model requests may still send context to your configured provider.

Questions, answered

Before you delegate the next task.

Agent choices, execution boundaries and what is available today.

An Agent works on a task inside a project. OpenLink connects the agent engine with its execution environment, tool activity and review surfaces.

Bring the next task into focus.

Start with the project. Follow the agent. Review the result.

One project. Every path to production.