Brev

Brev is a developer platform for accessing GPUs and deploying software on them. It connects multiple compute sources, simplifies SSH access, and offers Launchables for one-click software deployment. NVIDIA acquired Brev.dev in 2024; Nader Khalil announced the acquisition on July 16, and investor Nama Ventures followed with its announcement on July 20.[1][9][8]

Brev's GPU selection interface foregrounded the hardware developers requested. In his March 2026 account, Khalil recalled users asking for an A100 while conventional cloud provisioning buried that name behind forms and instance codes. His design question was: “what if the biggest piece of text was what the user’s asking for?” He designed the GPU graphics in Figma and implemented their animation by changing SVG styles through JavaScript with slow transitions.[1]

Brev's 2024 NVIDIA NGC integration paired cloud orchestration with a simple interface, browser-accessible notebooks, and direct SSH access through the Brev CLI. Its API provided a unified interface across public clouds, on-premises data centers, and private or hybrid clouds. Brev's open-source Verb container tool installed CUDA and Python and resolved dependencies; the CLI managed SSH keys without requiring users to handle IP addresses or PEM files. In their article, published July 1 and updated July 25, Nirmal Kumar Juluru and Khalil reported that the integration reduced a normally hours-long deployment process to two to three minutes.[5]

The same collaboration supplied a concrete fine-tuning example: Brev provisioned the GPU environment and installed the software stack, while NVIDIA NeMo supplied the training framework. The tutorial downloaded Mistral 7B, converted it to .nemo format, prepared PubMedQA data as JSONL, added LoRA adapters for parameter-efficient fine-tuning, and evaluated the resulting model with trainer.test(model). This demonstrated an infrastructure-to-training workflow without reporting a numerical accuracy improvement.[5]

Before this GPU-focused work, Brev supplied consistent development environments. That earlier product provides context for its later automation of GPU software setup.[7][5]

Nama's June 2022 announcement identified three founders and their roles at the time: Alecsander Fong, CTO; Ali Ahmed, COO; and Nader Khalil, CEO, all Y Combinator alumni with experience in Azure infrastructure and Workday developer tools. Nama managing partner Mohammed Alzubi described their decision to pivot from a previous startup to build Brev. Khalil later recalled starting Brev in his parents' garage.[7][9]

The 2022 product provided a ready development environment with one command while preserving developers' preferred tools and workflows. Its environments could be shared across projects and developers. In June 2022, Brev announced that it had raised an undisclosed pre-seed amount from Nama Ventures; Nama described the round as closed in its financing announcement.[7]

By 2023, Brev participated in the emerging AI developer community. It was among the projects represented at Agent Weekend, alongside AutoGPT, Codium, E2B, Roboflow, Ben's Bites, and smol. Latent Space also organized an August 26, 2023 open-AI event with Brev.dev and Langchain at its San Francisco headquarters, later listed as “Anything But Wrappers: Llama with Brev.dev.”[4][3][2]

Khalil recalled that Evan Conrad of SF Compute encouraged the team to embrace its small size rather than present itself as a larger company. Brev adopted a shaka logo and brought surfboards and palm trees to its GTC booth. Khalil's then-fiancée helped apply vinyl lettering to a surfboard at 1 a.m. before the event. The team also reused its GPU SVG artwork for foil-printed GPU gift cards made by a local San Francisco printing shop.[1]

In his July 2024 acquisition announcement, Khalil credited NVIDIA collaborators Michael Balint, Manuvir Das, Manish Harsh, and Julio C. Tapia. The accompanying video said the team would stay together and the product would retain its name. Looking back in March 2026, Khalil described NVIDIA as providing more resources to pursue GPU access and one-click deployment. He reported rapid growth inside and outside NVIDIA and work with partners, customers, and independent software vendors, while explicitly saying he lacked exact numbers.[9][1]

The team's work extended to DGX Spark remote access. Khalil recalled prioritizing easy SSH access in early hardware discussions; NVIDIA Sync simplified that connection, while Brev's CLI wrapped SSH. In March 2026, he described a CES-launched early-access feature that registered a home Spark in a Brev account for remote use like a cloud node. Networking and security work remained outstanding, so the early-access release allowed experimentation while the team addressed those constraints.[1]

Brev also served as compute infrastructure for agents. Khalil reported internal guidance to run agents in isolated Brev cloud VMs off NVIDIA's corporate network while security practices developed. He credited Alec with a Brev CLI rewrite, underway in March 2026, to discover available compute types and support provisioning through commands that could be piped together.[1]