Agent platforms
The substrates agents are built and run on: orchestration, tool and data connectors, memory, deployment and governance as one integrated surface. The factory, not the agent that comes off the line. You skip the plumbing and inherit the platform's model choices, pricing and limits. Click a card for its full spec.
The substrate teams build and run agents on, not the agent itself
Platforms sell the substrate rather than the agent. They provide the runtime, orchestration, evaluation and deployment surface that other teams build on, which puts them one layer below the agents in the rest of the catalogue.
Show more
The commercial logic is straightforward: agents are easy to prototype and hard to operate, so the durable business is in the operating layer. Expect the platform tier to absorb more of what teams currently hand-roll, particularly memory, tool governance and evaluation, which are the parts that get harder as a fleet grows.
What is an AI agent platform?
A managed service for building, running and monitoring agents. It supplies the orchestration, tool connections, memory, hosting and logging, so a team writes the agent’s logic instead of the infrastructure around it.
What is the difference between an agent platform and an agent framework?
A framework is a code library you build with and host yourself. A platform is a hosted environment that also runs the agent in production and gives you monitoring, access control and deployment. Some vendors sell both.
Should I build an AI agent or buy a platform?
Buy first if the value is in the agent’s job rather than its plumbing. Build when you need control over data residency, model choice or cost at scale. Check how easily prompts, tools and data export before committing, because moving off a platform later is real work.
What should I look for in an agent platform?
Which models it supports, what tools and data it can connect to, how it handles memory and long-running tasks, and what approval and audit controls it exposes. Tool connectivity increasingly runs through MCP, so check whether the platform supports it (Model Context Protocol).