> ## Documentation Index
>
> Fetch the complete documentation index at: https://lithi.ai/llms.txt
>
> Use this file to discover all available pages before exploring further.

---
title: Build reusable workflow profiles
description: >-
  A profile binds your team's defaults for quality, data policy, output and limits, and
  carries a version. A profile can only narrow account or key authority, never widen it.
route: /docs/compute/configure/profiles
page_id: docs-compute-configure-profiles
page_type: howTo
content_layer: guided documentation
surface_profile: guided_docs
audience: Teams standardizing how their Compute work is run
voice: D
reader_question: How do I set team defaults for Compute work without weakening our limits?
primary_action: Compare the quality presets
source_locale: en-US
source_status: APPROVED
source_version: docs-compute-configure-profiles-en-us-2026-09-05-v1
last_updated: '2026-09-05'
robots: index
claim_registry_pinned: true
claim_registry_resync: "npm run governed:tsx -- scripts/build-compute-public-claim-registry.ts --write"
---

## Before you start

You need one workflow your team already runs, and agreement on how it should be run by default. A profile records that agreement; it does not invent it.

You also need to know your account and key limits, because a profile is checked against them.

## What a profile binds

A profile is a named, versioned set of defaults. It carries four things.

- Quality: which preset the work runs at.
- Data policy: which data profile the input is allowed to be.
- Output: the format and the destination results are written to.
- Limits: the deadline and the maximum price a request may reach.

A profile is not a wire field of its own. When you name one on a request, it resolves into the request's policy, quality, output and limits objects.

## Choose the quality preset deliberately

Four presets exist, and each proves something different.

| Preset | What it proves | What it does not prove |
|---|---|---|
| Standard | The output matched the schema and passed deterministic validation | That the content is correct |
| Verified | An independent sample was checked | That every unchecked item is correct |
| High assurance | A higher share was checked, or checked by more than one independent reviewer | That the result is certified |
| Custom evaluation | The output met your own rubric or reference set | Anything your rubric does not test |

No preset is a correctness guarantee. Pick the one whose evidence you would actually act on.

## A profile narrows; it can never widen

This is the rule that makes profiles safe to share. Every value in a profile is checked against the account, project and key authority above it.

A profile may set a lower maximum price, a shorter deadline, a stricter data profile or a smaller output scope. It may not raise a ceiling, extend a deadline past the account rule, or admit a data profile the account rejects.

An override that would widen a higher-level rule is refused with a reason and a next action. Read that reason as intended behavior, not as a bug in your profile.

## Version every profile

Create a profile, then version it rather than editing it in place. Running work keeps referring to the version it was admitted with.

1. Plan the profile with its quality, policy, output and limit values.
2. Apply and read it back before anyone uses it.
3. Create a new version when a default changes.
4. Retire an old version once nothing references it.

Retiring is not deleting. Past results and receipts keep naming the version that produced them, so old work stays explainable.

## Confirm success

The profile reads back as `VERIFIED`, and a quote requested with the profile shows the limits and preset you expect.

Compare one quote made with the profile against one made without it. If the two look identical, the profile is not doing what you think.

## When a profile refuses

`REFUSED` on a widening override means the profile asked for more authority than the account allows. Lower the value, or change the account rule through its own approval.

`CONFLICT` means the profile moved since your plan was made. Inspect the current version, replan, and apply again.

[Compare the quality presets](/docs/compute/workflows/quality-presets)