What is an AI agent platform?
An AI agent platform is the layer that lets a team run, govern, and operate AI agents — not just a library to build one. It adds the user management, permissions, audit, and deployment story that a framework or runtime leaves out.
Framework vs. runtime vs. platform
Section titled “Framework vs. runtime vs. platform”These three terms are often used interchangeably, but they sit at different levels:
| What it is | Examples | What you still build yourself | |
|---|---|---|---|
| Framework | A code library for assembling agents | LangChain, CrewAI, AutoGen | UI, users, permissions, deployment |
| Runtime | An engine that runs an agent with tools | OpenClaw | Governance, multi-user, audit |
| Platform | The operational layer on top of a runtime | Pinchy | Mostly configuration, not code |
A framework gives you building blocks. A runtime gives you a working engine. A platform gives you the thing you can hand to a whole team.
What a platform provides
Section titled “What a platform provides”- Multi-user access — invite a team, with roles, not a single local user.
- Per-agent permissions — allow-listed tools so each agent can do only what it should.
- Audit trail — a verifiable record of every agent action.
- A chat interface and channels — somewhere for end users to actually reach the agents.
- Deployment — a defined way to run, update, and host the whole stack.
In Pinchy
Section titled “In Pinchy”Pinchy is an AI agent platform built on the OpenClaw runtime. OpenClaw provides the engine — model, tools, sessions, channels. Pinchy adds the enterprise layer on top: multi-user accounts, role-based access control, per-agent tool permissions, a signed audit trail, web and Telegram access, and a single Docker Compose deployment. See the Pinchy website and the Installation Guide.