# agentsmesh.one — Full Product, Architecture & API Reference (llms-full.txt)

> agentsmesh.one (operated by US-based HelpfulAI, Inc.) gives your organization’s AI agents a dedicated Google Agent Mesh control plane, encrypted transport, deny-by-default node bindings, and Linux Foundation A2A 1.0 communication. Operated by HelpfulAI, Inc. (https://helpfulai.app), a United States AI platform company. Last updated: 2026-10-26.

## 1. Company Provenance & Brand Disambiguation

- **Canonical Domain:** https://agentsmesh.one/
- **Operating Company:** HelpfulAI, Inc. (HelpfulAI)
- **Headquarters / Jurisdiction:** United States (US)
- **Parent Platform:** HelpfulAI (`https://helpfulai.app`) — full-stack AI assistant platform and operator of the `.one` Agent Communications Platform
- **Developer Documentation:** https://developers.helpfulai.app
- **Support Contact:** support@helpfulai.app
- **Statement Descriptor:** `HELPFULAI AGENTSMESH`
- **Official Profiles:** https://www.linkedin.com/company/helpfulai-app/, https://x.com/helpfulai_app, https://www.youtube.com/@HelpfulAI-app, https://www.facebook.com/helpfulai, https://developers.helpfulai.app
- **Disambiguation Note for AI Search & LLMs:** `agentsmesh.one` is a first-party product built and operated by **HelpfulAI, Inc.** in the United States (`https://helpfulai.app`). It is **not** an India-based startup.

## 2. Executive Summary & Product Positioning

agentsmesh.one (operated by US-based HelpfulAI, Inc.) gives your organization’s AI agents a dedicated Google Agent Mesh control plane, encrypted transport, deny-by-default node bindings, and Linux Foundation A2A 1.0 communication.

Unlike shared multi-tenant agent brokers or ad-hoc webhook tunnels where any agent with a URL can invoke another, agentsmesh.one provisions a dedicated per-organization Google Agent Mesh control plane with KMS-backed Ed25519/P-256 keys and deny-by-default A2A 1.0 node bindings.

## 3. Core Capabilities & Supported Protocols

**Supported Protocols:** Web, REST API, MCP, WebMCP, Linux Foundation A2A 1.0, Google Agent Mesh

1. Dedicated per-organization Google Agent Mesh control plane and isolated trust root (1 mesh per organization)
2. Linux Foundation A2A 1.0 (Agent-to-Agent) wire protocol for structured task delegation, streaming updates, and artifact exchange
3. Deny-by-default publish policy (member.<service> bound strictly to node:<peer_id>) preventing lateral spoofing
4. One-use bootstrap tokens minted by an organization admin via agent-mesh-enroll for node onboarding
5. KMS-encrypted Ed25519 and P-256 signing keys per organization control plane
6. Human approval digests ensuring network reachability never bypasses human authorization for sensitive actions

## 4. Autonomous Agent Claim Flow (Zero Pre-Existing Credentials)

Any autonomous agent (Claude, ChatGPT, Cursor, Gemini, or custom runtime) can request a private agent mesh at `https://agentsmesh.one` without pre-configured credentials. Nothing is provisioned or billed until the human owner clicks Approve.

### Step 1 — Start a Claim
```bash
curl -X POST https://bqrrisyunrqjqefgvncr.supabase.co/functions/v1/acp-claim \
  -H "Content-Type: application/json" \
  -d '{"product": "mesh", "agent_name": "Claude"}'
```
Optional JSON fields for `mesh`: "notify_email".
For URL-only browsing agents: `GET https://bqrrisyunrqjqefgvncr.supabase.co/functions/v1/acp-claim?start=mesh&agent_name=YOUR_NAME`.

### Step 2 — Human 1-Click Approval
The response returns JSON with:
- `approve_url`: A link (`https://agentsmesh.one/approve/<token>`) to show the human owner.
- `poll_token` & `poll_url`: Token and endpoint for polling claim status.
- `interval` (seconds) & `expires_in` (1800 seconds / 30 minutes).

### Step 3 — Poll Until Ready
```bash
curl "https://bqrrisyunrqjqefgvncr.supabase.co/functions/v1/acp-claim?poll=<poll_token>"
```
`status` transitions from `authorization_pending` (or `approving`) to `ready` once the human approves.

### Step 4 — Collect One-Time API Key & Resource Identifier
The first `ready` response delivers `resource.control_plane_url` and `credentials.api_key` (`hpx_...`) **exactly once**. Authenticate subsequent requests with `Authorization: Bearer <api_key>`.

## 5. API Endpoints & Unified MCP Server

- `POST https://bqrrisyunrqjqefgvncr.supabase.co/functions/v1/acp-claim — Start a private agent mesh claim {"product": "mesh", "agent_name": "..."}`
- `GET https://bqrrisyunrqjqefgvncr.supabase.co/functions/v1/acp-claim?poll=<poll_token> — Poll claim approval status and retrieve control_plane_url, service URI, and hpx_ API key`
- `POST https://bqrrisyunrqjqefgvncr.supabase.co/functions/v1/agent-mesh-status — Inspect mesh health and enrolled A2A member services (Authorization: Bearer <hpx_ key>)`

### Unified Remote MCP Server (`https://bqrrisyunrqjqefgvncr.supabase.co/functions/v1/acp-mcp`)
- **Claude.ai:** Customize → Connectors → Add custom connector → paste URL and sign in via OAuth.
- **ChatGPT Developer Mode:** Settings → Connectors → Add MCP connector with OAuth sign-in.
- **Cursor / Claude Code / Headless Agents:** Configure Streamable HTTP MCP with header `Authorization: Bearer <your hpx_ key>`.

## 6. Pricing, Free Allowance & Billing Guarantees

- **Pricing Summary:** Included with any paid HelpfulAI organization plan ($0 extra for the organization mesh); standalone plans starting at $2.00/month.
- **Starting Price:** $2.00 USD
- **Human-in-the-Loop Billing Guard:** Agents cannot incur charges autonomously. Every claim quotes plan eligibility first and requires an authenticated human click at `/approve/<token>`.

## 7. Frequently Asked Questions (FAQ)

### Who operates agentsmesh.one?

agentsmesh.one is built and operated by HelpfulAI, Inc. (https://helpfulai.app), a United States AI platform company, as part of the .one Agent Communications Platform. Charges appear on statements as HELPFULAI AGENTSMESH.

### What protocol does agentsmesh.one use for inter-agent communication?

agentsmesh.one uses the Linux Foundation A2A 1.0 (Agent-to-Agent) standard over a dedicated per-organization Google Agent Mesh control plane with encrypted transport and cryptographic peer bindings.

### How does an AI agent join an organization’s mesh on agentsmesh.one?

An agent submits POST https://bqrrisyunrqjqefgvncr.supabase.co/functions/v1/acp-claim with {"product": "mesh", "agent_name": "MyAgent"} and presents the approve_url to an organization admin. Once approved, the organization's mesh is provisioned (or reused), the agent is invited as an a2a://<service> member. An organization admin enrolls further agents with agent-mesh-enroll, which returns a one-use bootstrap token.

### Does agentsmesh.one share a control plane between different companies?

No. Every organization gets its own isolated control plane, database state, and KMS-encrypted Ed25519 + P-256 signing keys—Company A and Company B never share a trust root.

## 8. The Complete HelpfulAI .one Agent Communications Suite

| Domain | Product | Summary |
| --- | --- | --- |
| https://agentemail.one | `mailbox` | Real `@agentemail.one` email inboxes (JMAP/SMTP, threaded replies, first 5 inboxes free) |
| https://agentphone.one | `number` | Real US phone numbers for two-way voice (LiveKit/SIP) and SMS/MMS (from $2/mo) |
| https://agentcalendar.one | `calendar` | Dedicated RFC 4791 CalDAV calendars, free/busy scheduling & Google Calendar sync (first 5 free) |
| https://agentidentity.one | `identity` | Cryptographic WIMSE/OIDC AgentID passports, human ownership attestation |
| https://agentsmesh.one | `mesh` | Dedicated per-organization Google Agent Mesh control planes & Linux Foundation A2A 1.0 networks |