AI Infrastructure

AI systems are usually made of more than one model call. A request can move through an agent framework, a gateway, a sandbox, a workflow runtime, a hosted model API, or a self-hosted inference server before the user sees the final answer.

This section groups Parseable integrations by the part of the AI stack they observe. Use the framework guides when you want agent, model, and tool-call traces. Use the sandbox and runtime guides when you need visibility into execution environments and long-running workflows. Use the LLM and gateway guides when the main question is about model traffic, token usage, latency, and errors.

Agent frameworks

Agent frameworks usually carry the highest-level application context. They can show which agent ran, which prompt or task started the run, which model was called, which tool was used, and where the run failed or slowed down.

Coding assistants

Coding assistants emit agent-level telemetry from interactive development sessions. Use these guides to inspect model calls, token usage, tool execution, latency, errors, and session activity.

Sandboxes

Sandboxes show what happens when agents execute code and jobs in isolated environments. These guides cover sandbox activity, logs, traces, and metrics.

Agent runtimes

Agent runtimes carry work across steps, retries, and long-running workflows. Use these guides to follow execution and diagnose failures across the workflow lifecycle.

LLMs

LLM integrations help when you want to observe model-facing traffic directly. This can mean hosted provider APIs, SDK-level calls, or self-hosted inference runtimes such as vLLM.

Gateways

Gateways sit between applications and model providers. They are useful when teams want one place to manage routing, keys, budgets, policies, fallbacks, and model-provider telemetry.