What is VX Agents?
VX Agents is a platform for building AI agents that don't just talk — they do things in your real business systems, safely.
The connectivity layer between AI agents and your business systems.
You connect the tools your business already runs on — your CRM, booking system, payment provider, analytics, social channels, or your own internal APIs — and VX Agents turns them into actions an AI agent can perform on your behalf. Customers (or your team) chat in plain language; the agent looks up an order, reschedules a booking, pulls a report, replies to a comment, and answers with a real result.
It works from the built-in chat, an embeddable widget, and — for technical teams — from AI clients like Claude or ChatGPT through the Model Context Protocol (MCP).
The problem it solves
AI models are good at understanding a request. The hard part has always been letting them act on it against real systems without handing over the keys.
- Your business systems don't speak "AI tool call."
- Wiring each one up by hand is slow and brittle.
- Giving an LLM unscoped access to production is a non-starter.
VX Agents sits in the middle: it speaks to your systems on one side and to AI agents on the other, with permissions, authentication, and auditing built into the layer itself — not bolted on.
How it works
- Connect — Add a connection from the marketplace (Stripe, GitHub, Slack, Xero, Plaid, Google Analytics, Google Search Console, and more), connect a social account, or bring your own OpenAPI/GraphQL spec — including internal, legacy APIs.
- VX Agents generates tools — Each connection becomes a set of scoped, named actions the agent can use.
- The agent acts — In a conversation, the agent chains those actions to get real work done and replies with the outcome.
- Everything is governed — Credentials are stored encrypted and injected server-side, calls run behind an SSRF guard, and every action is audited under your organization, plan, and billing.
Because the tools live in the platform, the same agent is available in the web chat, in an embedded widget on your site, and as an MCP server for external AI clients — one setup, every surface.
Safety by construction
The thing that makes VX Agents more than a fancy HTTP client is that safety is part of the design, not a prompt hoping the model behaves:
- Scoped tools — an agent can only use what a connection actually exposes. A read-only connector (like the Verified Google Search Console template) simply has no write actions, so an agent cannot change your configuration even if asked.
- Server-side credentials — tokens never reach the browser or the model; they're injected on the server at call time.
- Guardrails baked in — SSRF protection, tool-count limits, content locks on sensitive templates, and full audit logs.
- Managed, Verified connectors — curated templates with OAuth2 and declared scopes, reviewed before they carry the Verified badge.
Safety isn't the agent promising to be careful — it's the tool surface simply not containing the dangerous verbs.
Who it's for
- Business owners who want customers to chat with their business systems — check an order, book a slot, get a quote — without building a bot from scratch.
- Technical teams who want to expose a CRM, booking, ERP, inventory, or internal API as secure MCP tools for ChatGPT, Claude, and customer messaging channels.
Under the hood
Perplexity — per agent or system-managed.
- Protocol: an MCP server per agent, so any MCP client can connect.
- Infra: containerized, with a controlled egress proxy for outbound calls.
Try it
There's a free self-serve plan. The fastest way to feel the idea is to connect one thing — a SaaS API, a social account, or your own spec — and watch an agent actually use it.
youtube.com
If you're working on the "AI agents meet real systems" problem too, I'd love to compare notes in the comments — especially on how you think about permissions and auditability once an agent can touch production.
Top comments (0)