Skip to main content
YOUR SERVER. YOUR AI WORKSPACE.v0.0.21.beta

FROM TO-DO TO DONE

From to-do.
To delivered.

Too many sources. Too many steps. Give xAgent the goal.
Turn research into briefs, sheets into reports, and meetings into next steps.

See it in actionStart deploying

Self-hosted · Multi-user workspaces · Policy-based approvals

THE REQUEST

Prepare a product launch kit from official sources.

3 official sources
xAgent / workspacePRODUCT INTERFACE

Actual execution record · Checked with follow-up instructions · For review

ACTUAL TASK ARTIFACTSThe result is in the files.
MDLaunch copy
MDFact check
HTMLLaunch page
Checked with follow-up instructions
01 / SET THE GOALStart with what you have
02 / MOVE WORK FORWARDDelegation, tools, and approvals
03 / KEEP THE RESULTFiles you can inspect and edit

GO BEYOND AN ANSWER.

Every step.
Toward a result.​

ILLUSTRATION · NOT A LIVE TASK
01/ 04

One request starts the work.

Start in a session, or use configured scheduled and external triggers.

GOAL + SOURCES + OUTPUT

START WITH THE OUTCOME

What would you hand over first?​

Pick a scenario. Explore every step.
See how a request becomes a deliverable.

DEMOINTERACTIVE EXAMPLERecorded real demo. No live AI calls.
YOUR REQUEST

One launch kit. Two sessions working together.

Use three official sources to prepare a launch kit for review. A checks facts; B creates an offline page and short copy. Collect, inspect, and revise the files without publishing them.

3 official sourcesDefined roles and deliverables

Review corrected the role split, the distinction between illustration and execution, and blanket approval claims. This is a checked run with follow-up instructions, not an unattended first pass.

DELIVERABLE SUMMARY.md
xAgent WORKSPACE / DEMO

xAgent product launch kit

Actual deliverables summarized · Checked with follow-up instructions

01 / Offline product page

launch.html covers the audience, capabilities, setup, and this collaboration. It has no registration form or external resources; links lead to official docs.

02 / Short launch copy

launch-copy.md contains 204 non-whitespace Chinese characters of main copy. It retains same-user, tool-configuration, and policy-dependent approval limits.

03 / Fact-check record

fact-check.md compares claims with the official sources read during the run, separating documentation from observed execution. This original snapshot still needs checking against current versions.

ACTUAL FILES · FOR REVIEW01
Download launch copy (Chinese)Explore session collaboration

Research and meeting examples are presets. The launch kit records an actual two-child-session workflow, checked with follow-up instructions rather than completed unattended in one pass. This page shows a record, not live AI.

Download the launch kitHTML page + launch copy + fact check · Original demo files
See the actual handoff and its limitsTwo child sessions · Three files

The coordinator created A for fact-checking and B for the page and copy. Both returned file references through collaboration events, after which the coordinator resumed reading and review. B received one follow-up to finish and return the copy; the final review also corrected role and approval claims.

01 / Actual collaboration record: creation and delegation
02 / Review and collect three files

This run used OpenAI-6.1-sol for coordination and Qwen3.8-27B for both child sessions. It validates this bounded handoff, not identical results across all models or complex tasks. External-action approval was not tested in this run.

The files were exported from product previews and preserve the sources checked during that run. Some capability descriptions still need checking against current versions, especially Word editing. This is not a current capability catalog or proof that the product download button was validated.

Original launch copyOriginal fact-check snapshotRead the capability guide

THE WORKSPACE BEHIND THE WORK

Progress you can follow.​

Explore the full manual

See the work. Keep the result.

Add requirements, inspect execution, and find files in the session. Complex tasks can use independent sessions, then bring their results together.

See how it works

Captured from the current Chinese demo environment on 2026-10-01. Listed capabilities are not all validated by this demo.

CAPABLE BY DESIGN. BOUNDED BY YOUR CHOICES.

Delegate the work. Keep a hand on the controls.​

Run it in your environment

Deploy on your server with separate user workspaces. Model and connector data flows depend on your configuration.

Read the details and limits

Review actions under your policy

Configure approval rules for actions such as sending or modifying. Inspect requests and choose whether they proceed.

Read the details and limits

Understand usage

Review personal model, token, and tool usage. Administrators can access broader usage statistics.

Read the details and limits

START WITH ONE REAL TASK

Try a task. Then make the call.​

Start with non-sensitive material.
Build confidence before expanding the scope.

01

Prepare the environment

Install the server, connect a model, and check workspace and security settings.

Deployment guide02

Complete one task

Bring a clear goal, sources, and output requirements. Start with a file you can inspect.

First-task guide03

Keep what works

After reviewing the result, turn the method into a reusable Skill or dedicated agent.

Create a reusable Skill
BETA

The free binary edition is for evaluation. It supports 2 users and 30 sessions; model and server costs are separate. See positioning and limits ↗

Before you begin

Does everyone need to install it?

No. Once an administrator deploys the server, users access it through the web or configured messaging channels. The server must stay online, and models and external services need setup.

Are these demos running live?

Not live. Research and meeting examples are presets. The launch kit, session handoff, and page screenshots come from an actual task checked with follow-up instructions. The animation illustrates the workflow; it is not a second-by-second replay. This page makes no model calls or uploads.

What exists today, and what is still planned?

Session notifications, collaboration events within the same user, and long-term memory exist today. Full project-level Agent Teams, further memory improvements, and knowledge-base features are still planned.

xAgent

Your next deliverable starts here.

Deploy xAgent