DataPacts

Enterprise Workflow RL Environments

Checked

At a glance

Enterprise workflow RL environments are simulated copies of business software, such as chat, CRM and knowledge-work tools, or of one company's own systems, in which AI agents practice office tasks and a checker scores the result; AI labs buy them to train general agents, and companies commission them to train their own.

Independent reference. Every buyer is covered on the same terms, every figure carries its source and check date, and no buyer pays to be listed or placed higher.

Request an independent introduction

Tell us what data your company holds. We reply with the buyers that fit and what their terms leave out. Your details reach a buyer only after you agree to a named introduction.

IndependentYour details stay with us until you agreeNot selling is a valid outcome

As of 2026-11-17, no lab names a purchase of one, and seven builders describe them on their own pages.

The environments sit within AI training data licensing as a data type, because a company's workflow records can be licensed to builders that turn them into replicas, tasks or evaluations.

Every figure below is the stating party's own claim with its date, and builders' statements about lab customers stay apart from what the labs say in filings. The page leaves out build tutorials and platform selection for company AI teams.

What Makes an RL Environment an Enterprise Workflow Environment

An enterprise workflow is the chain of steps staff complete across business software, and an RL environment becomes an enterprise workflow environment when it replicates that software or a company's own systems so an agent can work through the chain. Three neighbors fall outside the definition:

  • Screen-level replicas and task suites for operating a computer belong to RL environments for computer-use agents.
  • Repository and terminal tasks belong to the coding page.
  • Sandbox infrastructure that only hosts software supplies compute, not training material.

The table lists the four forms that builders' own pages describe, as of 2026-11-09, with the party that states each.

Form What is replicated Example on record Stated by Source
Replicas of popular business software named SaaS products "training gyms" replicating enterprise software such as Slack and Salesforce, for frontier labs Deeptune deeptune.com
Cloned enterprise environments enterprise software, products not named "pixel-perfect cloned enterprise environments" beside task factories for Finance and Legal Chakra Labs chakra.dev
Copy of one company's own systems the customer's systems, data, APIs and users simulation platform for testing, training and benchmarking enterprise AI agents Veris AI veris.ai
Knowledge-work environments and enterprise task sets domain workflows Westworld Finance Diligence Bench, 88 problems (2026-08-11); 4 enterprise environments in a 2,000+ task offering Halluminate; Scale AI halluminate.ai; scale.com/rlenvironments

No builder page on record names an ERP, ticketing-system or spreadsheet clone. The software named is Slack and Salesforce, and Fleet AI describes training gyms without naming the software they replicate, so a reader looking for those categories finds a gap, not a product.

The parent page explains what RL environments are and who buys them, and this page keeps to the business-software type.

Builders of Business-Software Replicas, With the Legal Entity on Record

Seven companies describe enterprise workflow environments on their own pages, listed alphabetically with the legal entity on record and checked November 9, 2026. The table gives the entity, the registry or footer source and what each builder's pages say about enterprise workflows.

Builder Entity and registry on record What its own pages say about enterprise workflows Checked Source
Chakra Labs Chakra Digital Labs, Inc. (SEC CIK 0002106724, Delaware, incorporated 2024, Brooklyn) cloned enterprise environments; task factories for Finance, Legal, Safety/RAI and computer use 2026-11-09 chakra.dev; sec.gov
Deeptune Deeptune, Inc. (SEC CIK 2110959, Delaware, New York) training gyms replicating enterprise software such as Slack and Salesforce, for frontier labs; Mercor announced on 2026-07-09 that it will acquire Deeptune, closing not confirmed as of 2026-11-17 2026-11-09 deeptune.com; mercor.com
Fleet AI Fleet AI, Inc. (terms and privacy notice; registry not opened) training gyms: high-fidelity simulation environments for agents 2026-11-09 fleetai.com
Halluminate Halluminate, Inc. (footer; no registry number on primary pages) benchmarks and RL environments for knowledge work, finance first (Westworld); Data Partnerships for company data and workflows 2026-11-09 halluminate.ai
Huzzle Labs Huzzle Ltd. (Companies House 13245967, incorporated 2021-03-04, London; Huzzle Labs is a division) coding, computer-use and enterprise-workflow RL environments for frontier labs; InsureBench 2026-11-09 labs.huzzle.com
Scale AI (RL Environments) Scale AI, Inc. (Delaware corporation per its Outlier terms) 2,000+ tasks, 4 enterprise environments, 870,000 artifacts 2026-11-09 scale.com/rlenvironments
Veris AI Veris Technologies Inc. (footer; no registry number on primary pages) a simulated copy of a company's systems, data, APIs and users for enterprise AI agents; priced per bench or annual license 2026-11-09 veris.ai

A builder is listed when its own page names enterprise software, enterprise environments or a company's systems, and the parent hub's grouping of five enterprise builders is the starting point.

Huzzle Labs and Scale AI join the list because their own pages name enterprise-workflow environments, and the same two also appear on the computer-use page, each described there only for its computer-use work.

The list is short and can change: a company moves in when its own page names the type, and out when that wording is removed. Companies whose identity is not confirmed are not described, and funding stays on the companies page: see RL environment companies with legal entity and funding.

Two Customers for Enterprise Replicas: AI Labs and Companies Training Their Own Agents

Enterprise replicas have two kinds of customer on record: AI labs that train general agents on copies of popular software, and companies that pay for a copy of their own systems to train or test their own agents.

On the lab side, Deeptune states that it builds for frontier labs, and Halluminate states that four of the five leading closed-source U.S. AI labs are customers, without naming them. Huzzle Labs shows a Meta Superintelligence Labs testimonial on its own site.

Anthropic's S-1 of October 26, 2026 says its training data includes datasets licensed or commissioned from third parties and that it suspended higher-risk RL environments, and it names no environment type.

The table orders the statements by date, as of 2026-11-17, and shows what each does not establish.

Date Party Statement Stated by What it does not show Source
2026-03-16 companies buying through Google Cloud Marketplace Veris AI available on Google Cloud Marketplace Veris AI named customers or volumes veris.ai
2026-07-06 (MSA last updated) frontier labs and enterprises custom data and environments under a public MSA; undisputed amounts due within 30 days of monthly invoices Proximal named customers or prices proximal.ai/msa
2026-10-01 four of the five leading closed-source U.S. AI labs, unnamed customers of Halluminate Halluminate which labs, which environments, any price halluminate.ai
2026-10-26 Anthropic training data includes datasets licensed or commissioned from third parties Anthropic, PBC (S-1) which datasets, or any enterprise environment sec.gov
2026-10-26 Anthropic higher-risk RL environments suspended after 2026 incidents in pre-release cybersecurity evaluations Anthropic, PBC (S-1) environment types or builders sec.gov
2026-11-03 Fireworks AI customers Foundry: custom environments from the customer's own sessions, tools and test cases, early access by request Abundant named customers, prices or license terms abundant.ai
checked 2026-11-09 Meta Superintelligence Labs testimonial on Huzzle Labs' site Huzzle Labs a contract, scope or price labs.huzzle.com

Only the Anthropic rows come from a buyer's own filing; the others are builders' pages, and no lab names an enterprise environment purchase.

Companies appear as customers in builders' own terms, and the record is thinner than on the lab side because only Veris AI and Proximal publish how a company buys.

Veris AI sells a simulated copy of a company's systems per bench or under an annual license, Proximal sells to frontier labs and enterprises under its public MSA, Bespoke Labs states that it builds RL environments for frontier labs and enterprises, and Prime Intellect sells compute and training to enterprises.

Abundant's Foundry, announced with Fireworks AI on 2026-11-03, starts from a Fireworks AI customer's user sessions, tools and golden test cases, with early access by request and no published license terms. Veris AI states Google Cloud Marketplace availability from 2026-03-16 but names no customers.

Two price statements exist, both on 2026-01-12 from Epoch AI: one interviewee puts a high-fidelity replica such as Slack at around USD 300,000, and SemiAnalysis via Epoch AI puts a website replica (a UI gym) at about USD 20,000.

Neither is a market price. The page on public price points for RL environments collects the rest.

What a Company Can License Into These Environments: Workflow Records, Not Customer Identities

A company can license workflow records, the history of how work moved through its chat, CRM, ticketing and document tools, to builders that turn them into environments or tasks, and every program on record states that identities are removed or that the seller keeps control.

Builders acquire such records to make replicas realistic, because a copy of Slack or Salesforce with invented content behaves differently from one seeded with real task histories. Programs and builders overlap, so the table covers both.

The table lists seven programs alphabetically by operator, as of 2026-11-09, with the personal-data handling and terms each states.

Program What it asks for Personal data and control as stated Terms stated Checked Source
Abundant Foundry and "License your data" user sessions, tools and golden test cases; datasets offered for licensing not published not published 2026-11-09 abundant.ai
Collinear AI paid data licensing private repos with PRs and tickets, Slack, Gmail, Jira, Salesforce, HubSpot, Figma, database dumps export and anonymization before payment; seller keeps ownership; one-time snapshot no amounts; paid per contract after delivery; teams of about 50+ 2026-11-09 collinear.ai
Halluminate Data Partnerships company data and workflows, types not specified form asks for no files and no confidential information payout and license not published 2026-11-09 halluminate.ai
Mercor Data Monetization Slack, Gmail, Salesforce, GitHub, Notion, Confluence and 100+ tools automated extraction and de-identification; "No AI lab buyer or third-party vendor ever sees your raw data" no amounts; wire typically 2 to 4 weeks after signing 2026-11-09 mercor.com/data
Ooak Data partner program Slack, Gmail, Notion, Jira, SharePoint, Teams in-house anonymization; raw archive deleted within 30 days of delivery; training use only "Earn up to $1M" (Ooak's claim); exclusivity not stated 2026-11-09 ooakdata.com
Scale AI Data Partnerships Salesforce, HubSpot, Workday, ServiceNow, SAP, Zendesk, Jira, chat, email anonymizes inside the company's environment before extraction "$10K to $1M+ illustrative value" (Scale's claim) 2026-11-09 scale.com/data-partnership
Turing Project Lazarus Slack, Teams, Jira, Linear, Salesforce, HubSpot, Zendesk, Intercom, code Turing covers de-identification; ownership does not change hands revenue share, nothing upfront, also paid when Turing builds RL environments on the record; license to Turing exclusive for the first 6 months (checked 2026-11-17) 2026-11-09 lazarus.turing.com

Records programs and environment builders are listed together because builders on record acquire the same workflow records; no row is a recommendation.

Customer and contact identities are not what these programs state they want. Ooak Data anonymizes in-house, Collinear AI exports and anonymizes before payment, and Mercor states that no lab buyer sees raw data. Contact data inside CRM records carries its own privacy rules, set out in selling CRM records for AI training.

Customer contracts can rule records out: Mercor's manufacturing listing is marked "not a fit" for data restricted by customer contracts (checked 2026-11-09). Environments are also not guaranteed clean, because DataVendor states that it cleans purchased files of personal information within 2 business days but does not clean tasksets and environments.

Licensing Workflow Records to an Environment Builder: Benefits, Risks, Downsides and Alternatives

Licensing workflow records to an environment builder can earn money from history the company already holds, but the environments built on it usually belong to the builder or the lab.

Benefits

  • Collinear AI states that the seller keeps ownership of the records and is paid per contract after delivery (checked 2026-11-09).
  • Turing states a revenue share that also applies when Turing builds RL environments on the record.
  • Ooak Data states that it deletes the raw archive within 30 days of delivery.

Risks

  • Builders can own the finished work: Proximal's public MSA states that Proximal owns Deliverables and may reuse them absent customer confidential information (s.8.1, last updated 2026-07-06), and rights in cloned software and seed data are covered on their own page.
  • An acquisition changes the counterparty, as with Mercor's announced acquisition of Deeptune.
  • Personal data left in an environment can travel with it.

Downsides

  • No environment builder publishes a payout for environment use, and Turing's split is not published.
  • No lab names its purchases, so a seller cannot see who the end customer is.

Alternatives

  • License the same records to a records program that pays for operating data, such as Scale AI's Data Partnerships.
  • License agent trajectories from the company's own agents: micro1 states it licenses qualifying trajectories for USD 1 to 2 million a year, with the company's approval (checked 2026-11-09).
  • Commission an environment of the company's own systems for its own agents, as Veris AI offers.
  • Not license at all, which leaves the records, and their risks, with the company.

These lists do not weigh one builder against another, and the right choice depends on the records, the customer contracts and the company's own tolerance for reuse. For the full process and the not-selling option, see how to sell or license company data to AI companies.

Where Enterprise Replicas Meet Computer Use, Rights, Grading and Workflow Data

This page covers environments that replicate business software and company systems; screen-level replicas, rights, grading, selling and each workflow data type have their own pages.

Frequently Asked Questions

Which business software do enterprise RL environments replicate?

The software named on builders' own pages is Slack and Salesforce, which Deeptune states its training gyms replicate; Chakra Labs describes cloned enterprise environments without naming products. No builder page on record names an ERP, ticketing-system or spreadsheet clone as of 2026-11-17.

Can a company have an RL environment built from its own systems?

Yes: Veris AI sells a simulated copy of a company's systems, data, APIs and users per bench or under an annual license, and Abundant's Foundry starts from a customer's sessions and tools. Proximal sells custom environments under a public MSA.

Do enterprise workflow environments contain customers' personal data?

They can, because DataVendor states that it does not clean tasksets and environments of personal information, while builders that buy records, such as Ooak Data, state that they anonymize before use.

What each treatment achieves legally is explained under anonymization versus pseudonymization.

Do AI labs name the enterprise environments they buy?

No lab names one on record as of November 17, 2026: Halluminate counts four of the five leading closed-source U.S. labs as customers without naming them. Anthropic's S-1 of 2026-10-26 names no environment type.