---
title: Compute tools
description: >-
  Heavy scientific compute such as folding, docking, ADMET, generative design
  and omics, which runs as tracked jobs, produces artifacts, and uses credits.
tags:
  - concepts
  - tools
  - jobs
  - reference
---

Compute tools are the platform's heavy scientific compute: folding, docking, ADMET prediction, generative design, and omics analysis. Each runs as a **tracked job** on hosted compute — a GPU when the model needs one, CPU otherwise — and writes artifacts you can revisit.

When the orchestrator calls one, a progress card steps through `queued → running → completed`, and the run consumes credits for the compute it uses. This page is the front door to that catalog, grouped by capability.

> **These run as jobs**

> A compute tool does not answer inline. It dispatches a job you can watch, cancel, and come back to, and its result is backed by a `jobs` record rather than pasted into the conversation. See [How tools run](/docs/how-tools-run) for how this execution class differs from an instant lookup or a sandbox script.

## The catalog by capability

Each card below links a representative tool in that sub-area — follow it, and the sibling links on the tool page reach the rest of the group.

## How these run

Compute tools are one of the platform's execution classes. Because they run on dedicated compute as tracked jobs, they carry the longest latency, produce artifacts that persist, and bill against your plan's credits.

- [How tools run](/docs/how-tools-run) — the execution-class explainer: what makes a tool a job versus an instant lookup, and where each output lands.
- [Plans and tiers](/docs/plans-and-tiers) — how credits meter compute and which tiers unlock which model families and tools.

## Read next
