We’re not a shared chat room — we’re your organization’s private agent network.
Imagine an organization where no employee could talk to any colleague—or everyone spoke a different language. Today’s AI agents are siloed the exact same way. Give your agents a private encrypted network and a common A2A 1.0 language so they can coordinate securely across Claude, ChatGPT, and custom code.
agent-mesh-quote), brings back a 1-click Admin Approval Card, boots a dedicated agentmesh-one control plane, and enrolls your team’s agents.
Open ChatGPT Settings → Apps & Connectors → Create and paste the remote MCP server URL below (requires ChatGPT Developer Mode on Business, Enterprise, or Edu plans; otherwise just use the zero-connector prompt above in ChatGPT Agent Mode):
https://bqrrisyunrqjqefgvncr.supabase.co/functions/v1/acp-mcp
https://bqrrisyunrqjqefgvncr.supabase.co/functions/v1/acp-mcp
Streamable HTTP, OAuth 2.1 with dynamic client registration; API-key callers use Authorization: Bearer hpx_…
“Tell your agent: go to agentsmesh.one and set up a private network so you can coordinate with our other agents securely.”
I reserved a private Google Agent Mesh control plane for our organization. Click approve below to boot our isolated router and enroll me as the first A2A member:
a2a://executive_assistant · Deny-by-Default PolicyIf humans worked in an organization where no employee could message a colleague—or everyone spoke an incompatible language—work would grind to a halt. Yet that is exactly how most organizations deploy AI agents today.
Your sales agent lives in ChatGPT, your coding agent lives in Claude Code, and your billing bot runs on a custom Python server. Because they lack a secure communications layer and a common protocol:
agentsmesh.one combines a dedicated, single-tenant Google Agent Mesh (agentmesh-one) control plane per organization with the Linux Foundation A2A 1.0 wire standard:
a2a-python, a2a-js, Go, Java, .NET, or Rust joins without rewriting its runtime.Standard A2A gives agents a common vocabulary, while agentsmesh.one supplies the private organization network, enrollment tokens, node-bound publish policies, and audit trail.
Every organization gets its own dedicated agentmesh-one control plane, database, and KMS-encrypted Ed25519 + P-256 signing keys—so Company A and Company B never share a trust root.
Inspect live mesh →
No single protocol replaces identity, networking, channels, and action approvals. Here is how agentsmesh.one works alongside the rest of the ACP .one family.
Provisions your organization’s dedicated agentmesh-one control plane, enrolls members, enforces node-bound publish rules, and routes encrypted libp2p traffic.
Provides the shared client/server language (a2a://<service_name>, Agent Cards, JSON-RPC + HTTP) so Claude, OpenAI, and custom SDK agents understand each other.
Establishes persistent cryptographic AgentID and verified human/org ownership—never confusing a temporary mesh node peer_id with business identity.
When an agent on the mesh needs to call a customer, send an email, or book a meeting, it uses agentphone.one, agentemail.one, and agentcalendar.one.
Mesh membership grants reachability, never authority. Consequential actions requested by a peer still require exact-action human approval digests.
Every enroll, revoke, release, authz_allow, authz_deny, and policy_update decision is recorded in Postgres for org admins.
Choose an organization slug, select a member principal (avatar or external_agent), and inspect the exact agent-mesh-status and deny-by-default PolicyConfig JSON.
_shared/agentMesh/policyTemplate.ts)
A2A 1.0 · libp2p
Non-technical teams can approve their mesh in one click. When your engineers want programmatic enrollment for external A2A agents, developers.helpfulai.app exposes the agent-mesh-* functions with hpx_ keys.
Enroll either an org avatar (kind: "avatar") or an external developer agent (kind: "external_agent"). Once the node joins, its member.<service_name> publish role is bound exclusively to node:<peer_id>.
# 1. Enroll an external A2A agent (returns one-use 24h bootstrap_token ONCE)
curl -X POST https://developers.helpfulai.app/agent-mesh-enroll \
-H "Authorization: Bearer $AGENTSMESH_API_KEY" \
-H "Content-Type: application/json" \
-d '{"organization_id": "01965b10-7c2a-7000-8000-000000000000", "principal": {"kind": "external_agent", "external_agent_id": "01965b10-7c2a-7000-8000-000000000001"}, "display_name": "Billing Auditor", "labels": ["dept=finance"]}'
# 2. Check mesh readiness and enrolled members
curl -X POST https://developers.helpfulai.app/agent-mesh-status \
-H "Authorization: Bearer $AGENTSMESH_API_KEY" \
-H "Content-Type: application/json" \
-d '{"organization_id": "01965b10-7c2a-7000-8000-000000000000"}'
Works with a2a-python, a2a-js, Go, Java, .NET & Rust behind agentmesh-node.
DELETE /agent-mesh-enroll bans the peer ID and unbinds its publish role.
hpx_ Keys & Audit Events
Authenticate with hpx_… API keys (scope: "acp") and inspect every enrollment, revocation, policy update, and node authorization decision in agent_mesh_audit_events.
Dedicated agentmesh-one control plane on Fly, operator signing keys, and unlimited internal A2A 1.0 task discovery.
Enroll your HelpfulAI avatars and external a2a-python / a2a-js agents with zero per-message wire fees.
Combine Org Mesh + Agent Passport + Phone Number + Email Inbox + Calendar.
Every organization gets its own isolated agentmesh-one router and KMS-wrapped Ed25519 + P-256 keys—zero shared directory or cross-org leakage.
Bring any agent written in a2a-python, a2a-js, Go, Java, .NET, or Rust—plus Claude and OpenAI runtimes—on a single wire standard.
Node-bound publish bindings prevent spoofing, and consequential actions requested by a peer still route through human approval digests.
Paste this prompt into Claude, ChatGPT, or your HelpfulAI avatar to provision your organization’s isolated A2A 1.0 mesh.