Alice, formerly ActiveFence, sells AI safety and security training datasets, agentic RL environments, and expert-led red teaming and evaluations to foundation model labs; it also sells WonderSuite guardrail software to enterprises and content moderation to online platforms.
What they offerTraining data, RL environments, Evals and benchmarks, Human data
Type of workSecurity, Computer use, Browser, Coding, Enterprise workflows
Data sourcesExpert or contractor-created, Synthetic / generated, Product replica or partner app
Archal sells resettable API sandboxes that mimic third-party software services, giving AI developers and software teams repeatable environments to test, evaluate, and train agents without using live accounts.
What they offerRL environments, Evals and benchmarks, RL infrastructure
Type of workMulti-domain
Data sourcesProduct replica or partner app, Synthetic / generated
Builds simulated, workflow-based environments and benchmarks, plus a runtime and managed post-training service, for AI labs and developers training and evaluating agents.
What they offerRL environments, Evals and benchmarks, RL infrastructure
Chakra Labs builds expert-authored tasks, graders and computer-use environments for training and evaluating AI models. Its off-the-shelf and bespoke task sets cover finance, legal work, responsible AI and computer use.
What they offerRL environments, Training data, Evals and benchmarks
Type of workComputer use, Enterprise workflows, Finance, Legal, Security
Data sourcesProduct replica or partner app, Expert or contractor-created
Deeptune builds simulated workplace-software environments, bundled with tasks, datasets and infrastructure, where frontier AI labs train agents to carry out multi-step digital work. Mercor acquired Deeptune in July 2026.
G2i supplies coding post-training data to AI labs (RL environments, evals, RLHF preference data, SFT datasets, scrubbed seed data) through managed projects or contract engineers placed on lab teams. It also runs an engineer hiring platform and tech conferences.
What they offerRL environments, Human data, Training data, Evals and benchmarks
Type of workCoding, Security, Enterprise workflows
Data sourcesExpert or contractor-created, Product replica or partner app
Matrices sells realistic simulated computer-use environments to AI labs, where multimodal agents practice browser and workplace tasks through reinforcement learning without real-world side effects.
What they offerRL environments, Simulation
Type of workComputer use, Browser, Enterprise workflows
ReadySolve licenses MartechBench, an enterprise marketing-workflow environment for reinforcement learning and evaluation. Agents operate connected business systems with deterministic task verification; buyers receive runtime materials and can commission model benchmarks.
What they offerRL environments, Evals and benchmarks
Type of workEnterprise workflows
Data sourcesExpert or contractor-created, Product replica or partner app
Scale sells AI labs and businesses custom training data, human feedback, annotation, RL environments, and model evaluations, alongside software to build, test, and deploy AI agents and applications.
What they offerRL environments, Training data, Evals and benchmarks, Human data, Data annotation
Type of workMulti-domain, Coding, Computer use, Browser, Enterprise workflows, Finance
Data sourcesSynthetic / generated, Product replica or partner app
Sticker builds reinforcement learning environments and benchmarks for training and evaluating AI agents on procurement work, including backorders, purchasing policies and mismatched invoices, for teams post-training models. It previously built and ran procurement infrastructure for businesses.
What they offerRL environments, Evals and benchmarks
Vibrant Labs builds RL-ready simulated environments, training tasks, and benchmarks for AI labs and developers working on long-horizon agents, including agents that use browsers, computers, or enterprise tools.
What they offerRL environments, Training data, Evals and benchmarks
Type of workMulti-domain, Computer use, Browser, Enterprise workflows
Data sourcesSynthetic / generated, Product replica or partner app
Last checked
Initial directory snapshot · exported Nov 4, 2026. Original source check dates are retained. Automatic monitoring is not active yet. How we check information