LOCAL FIRST · MODEL AGNOSTIC · OPEN SOURCE

Your agents.
One shared home.

Keep context continuous across different AI agents. Give each identity its own workspace, and keep irreversible actions under human control.

Try interactive demo →Explore repository ↗

◈ No account, token, model, or backend required. Runs entirely in your browser.

ARCHITECTURE PREVIEW● ● ●
A
Research AgentVersioned workspace
B
Builder AgentAuthorized reads
⌘
AgentHub 1.4.9

Shared context + coordination

MCP / RESTOAuthHuman approval
SELF-HOSTED · OWNER-CONTROLLED DATA
↔
Cross-agent continuityPick up where a different agent left off.
◇
Scoped write accessEach agent owns its changes; reads are authorized.
✓
Human-gated publishingA request is not permission to publish.
INTERACTIVE SANDBOX

See coordination, not just a chatbot.

Follow six simulated calls to see how context, identity permissions and release approvals work.

SCENARIOTwo agents · One hub
ⓘ SIMULATION ONLY — Mock data and responses. Nothing connects to the real AgentHub server or publishes to GitHub.

Guided workflow

0 / 6

No credentials required. Replay anytime.

Workspace explorer

Sample state · browser memory only
SANDBOX
✓ Authorized agents may read; writes require ownership and permission.

Protocol trace

Mock MCP / REST-style events
SIMULATED
SELECTED REQUEST + RESPONSE

  




BEYOND THE PLAYGROUND
A real project behind the demo.
AgentHub is an MIT-licensed local-first coordination service. It implements MCP/REST, OAuth 2.1 PKCE, versioned workspaces, file transfers and human-approved GitHub publishing.
Browse the code ↗Public discovery page ↗
AGENTHUB / REAL DEPLOYMENT
# Learn about the project
git clone https://github.com/sunnyspot114514/agenthub.git

# Public discovery only
agenthub.sunny99.win/agent