Runhouse
AcquiredNew York, US · Founded 2022 · 3 known investors
Find your way into Runhouse
Sign up to see every warm intro you have to Runhouse
- Paths you didn't know you had: your email and LinkedIn already hold routes to the Runhouse team. We find them for you
- 2nd- and 3rd-degree connections: the friend-of-a-friend routes that take hours to manually find through your inbox or LinkedIn
- Answers now, not in days: momentum is everything in a raise. Skip asking around whether someone knows someone
Days of research, done the moment you sign in, with every route ranked by how warm it is.
Runhouse built Kubetorch, a Kubernetes-native distributed framework for enterprise ML and reinforcement learning post-training.
Also known as Kubetorch
Founders & leadership
Investors · 3
Reported raises · per SEC filings
Form D private placements$4.9M disclosed across 1 of 2 rounds · 2023–2025
▶$4.9MraisedJun 2023 · 26 investors · Other TechnologyRule 506(b)
- Joshua LewittesExecutive Officer, Director
- Don GreenbergExecutive Officer, Director
- Offering amount
- $5.6M
- Amount sold
- $4.9M
- First sale
- Jun 2023
- Incorporated
- Corporation, Delaware, 2022
- Federal exemptions
- 06b
Source: SEC EDGAR Form D. Amounts as filed; amended filings shown once at their latest values.
Company profile
researched Aug 2026Runhouse developed infrastructure software for machine learning teams, centered on the principle that application code should drive infrastructure rather than the reverse. Its platform provided compute and data tooling that let teams program, share and observe infrastructure remotely, and was positioned for enterprises and AI-native teams moving AI post-training from research into production.
The company's flagship product, Kubetorch, is a distributed framework for Kubernetes purpose-built for enterprise machine learning and reinforcement learning workloads. It allows ordinary Python code to run on cloud hardware with fast iteration, reproducibility and scale, abstracting Kubernetes interactions behind Pythonic APIs for researchers while remaining Kubernetes-native for platform teams, so it integrates with existing observability and management stacks. Kubetorch can be deployed inside a customer VPC in roughly 15 minutes, across on-premises clusters, cloud GPUs or multiple Kubernetes clusters. Its interface is described as general-purpose and unopinionated, supporting programmatic fault tolerance and faster iteration and debugging of ML applications.
Runhouse addressed workloads that combine distributed training, inference and evaluation — heterogeneous patterns that reinforcement learning requires and that platforms such as SageMaker, Vertex AI and Kubeflow were not designed to handle. The company also targeted the slow research feedback loop caused by long CI pipelines and worker restarts, aiming to unify interactive experimentation with scalable, reproducible production execution.
Founding story
Runhouse was founded by Donny Greenberg and Josh Lewittes. Greenberg led product for PyTorch at Meta, and previously worked at IBM as a tech lead in quantum computing research and at Google as a product manager in large-scale ad infrastructure. Lewittes built fraud detection models at SecuredTouch, backend ML systems at KPMG, and worked as a machine learning engineer at Gloat, later serving as Runhouse's CTO. Both encountered the same recurring bottleneck — infrastructure slowing iteration and delaying the path from model ideas to production — which shaped the company's focus on letting enterprise teams iterate on models as easily as running code locally.
Business model
Enterprise software: Runhouse offered Kubetorch for deployment within customer environments (on-premises clusters, cloud GPUs, or multiple Kubernetes clusters) inside the customer's VPC, serving Fortune 500 companies, frontier AI labs and AI-native startups.
Traction
Kubetorch was reported to be powering production deployments at Fortune 500 companies, frontier AI labs and AI-native startups, with those teams using proprietary and live agentic user-interaction data for continuous model improvement. The product was in open beta as of October 2025.
Latest developments
In April 2026, Calcalist reported that Anthropic had acquired Runhouse in a deal conducted primarily in Anthropic shares and completed at a relatively modest valuation, with the transaction closing several weeks before disclosure. Investor Guy Fighel of Hetz Ventures described it as an acqui-hire whose primary value lay in the founders and team rather than intellectual property or customer base, and noted that the share-based consideration had appreciated since closing. Signalbase reported the same acquisition, for an undisclosed amount, framing it as a move to strengthen Anthropic's internal ML development and compute infrastructure.
▸Full profile — market position, technology, go-to-market, geography, history, risks & controversies
Market position
Runhouse positioned itself against incumbent ML platforms including AWS SageMaker, Google Vertex AI and open-source Kubeflow, arguing these were insufficient for the speed, customization and fault tolerance required by modern ML and reinforcement learning workloads, and against ML notebooks that cannot handle scale or workload heterogeneity. Its thesis was that competitive differentiation in AI is shifting from pretraining, dominated by frontier labs, toward post-training on proprietary enterprise data.
Kubetorch combined interactive, fast iteration with scalable and reproducible execution in one system, ran regular Python code on distributed cloud hardware, abstracted Kubernetes behind Pythonic APIs while remaining Kubernetes-native for platform teams, supported programmatic fault tolerance, and could be installed entirely within a customer VPC in about 15 minutes across on-premises, cloud GPU and multi-cluster environments.
Technology
A Kubernetes-native distributed framework for ML and reinforcement learning workloads, exposing a Python-based compute interface for training, inference and evaluation with reproducibility, programmatic fault tolerance and rapid iteration and debugging, deployable into existing cloud environments and technology stacks.
Go-to-market
Product-led access via open beta sign-up at run.house alongside direct enterprise deployment inside customer VPCs; investor-led announcement of the Seed round highlighted existing production usage at large enterprises and AI labs.
Enterprise ML, data science and platform teams, including Fortune 500 companies, frontier AI labs and AI-native startups; investor commentary cites applicability to domains such as law, finance, healthcare, life sciences and mechanical engineering.
Geography
Fully distributed with no central office; co-founder Donny Greenberg was based in New York and CTO Josh Lewittes in Israel, while one directory listing records San Francisco, United States as headquarters.
History
Reported founding dates differ between sources: one profile lists 2022, while Calcalist reports the company was founded in 2023. The company operated as a fully distributed team of about 10 people with no central office; one directory lists San Francisco as headquarters, while Calcalist places co-founder Donny Greenberg in New York and CTO Josh Lewittes in Israel. Runhouse raised a $5M Seed round announced in October 2025 and was in open beta at that time. Anthropic acquired the company in a primarily share-based acqui-hire that closed in early 2026 and was disclosed in late April 2026.
Risks & controversies
Sources report inconsistent basic company data, including founding year (2022 versus 2023), headcount (4, 6 or approximately 10) and headquarters location. Revenue and valuation figures appearing on one directory profile are that site's own estimates derived from comparable companies, with stated confidence levels of 64% and 10% respectively, and are not company-reported.
Compiled by commissioned research from 5 cited public sources — announcements, filings, and press listed under research sources below.
Key figures
latest reportedCompany-reported or press-reported figures, each dated to when it was claimed — not independently audited.
Timeline · 2
launches, deals, and filingsAnthropic acquired Runhouse in a deal conducted primarily in Anthropic shares at a relatively modest valuation. The transaction closed several weeks before being disclosed and is characterized as an acqui-hire, with value attributed primarily to the founders and team rather than intellectual property or customer base.
Work-Bench led Runhouse's $5M Seed round, with participation from Hetz Ventures and Fathom Capital. At the time of the announcement Runhouse was in open beta.
$5M source ↗
Dated company events from announcements, filings, and press; legal rows summarize public dockets and regulator releases.
▸Research sources · 5
primary sources listed
- Runhouse — $5M Raised, Investors, Team & Alternativesstartuphub.ai · web
5 public sources were cited for this profile; the first-party ones are listed here.
Frequently asked questions
- What does Runhouse do?
- Runhouse built Kubetorch, a Kubernetes-native distributed framework for enterprise ML and reinforcement learning post-training.
- Who are Runhouse's investors?
- Runhouse's investors include Work-Bench, Fathom Capital, Hetz Ventures.
- How much funding has Runhouse raised?
- Runhouse has disclosed $4.9M raised across 1 of its 2 known rounds.
- Where is Runhouse headquartered?
- Runhouse is headquartered in New York, US.


