# What is a Runtime?

> inference shell is an agent runtime; the infrastructure layer that executes your agent code with built-in solutions for the hard operational problems.

**URL:** https://inference.sh/docs/runtime/overview
**Last updated:** 2026-09-18

---

inference shell is an **agent runtime**; the infrastructure layer that executes your agent code with built-in solutions for the hard operational problems.

## runtime vs framework

a framework gives you building blocks. you import a library, define your agent logic, and figure out where to deploy it yourself.

a runtime is infrastructure that executes your agent with specific guarantees about:

- **durability**: state persists across failures and restarts
- **observability**: every decision and action is captured automatically
- **tool orchestration**: managed integrations with authentication handled
- **human oversight**: approval gates for consequential actions

the distinction matters because agent workloads have unique characteristics that general-purpose infrastructure handles poorly.

## why agents need a runtime

agents are different from traditional software:

- they make long sequences of decisions punctuated by external calls
- they need to maintain conversation context across many interactions
- they call tools that might fail, timeout, or return unexpected results
- they run for unpredictable durations; sometimes seconds, sometimes hours
- they need human oversight for sensitive actions

standard web frameworks assume request-response cycles measured in milliseconds. container orchestration assumes stateless workloads. message queues assume independent jobs. agent workloads fit none of these patterns well.

## what the runtime handles

when you run agents on inference shell, you get:

| capability | what it does |
|------------|--------------|
| [durable execution](/docs/runtime/durable-execution) | state checkpoints after each step; resume on failure |
| [observability](/docs/runtime/observability) | automatic tracing of every decision and tool call |
| [human-in-the-loop](/docs/runtime/human-in-the-loop) | approval gates for consequential actions |
| [tool orchestration](/docs/runtime/tool-orchestration) | integrations with managed authentication |

you write agent logic. the runtime handles operations.

## learn more

- [durable execution](/docs/runtime/durable-execution): how agents survive failures
- [observability](/docs/runtime/observability): visibility into agent behavior  
- [human-in-the-loop](/docs/runtime/human-in-the-loop): approval gates for actions
- [tool orchestration](/docs/runtime/tool-orchestration): managed integrations

or dive into the [deep dive on agent runtimes](https://inference.sh/ai-agent-runtime).