Skip to main content

​
What is Buildfunctions?

Buildfunctions provides the traditional benefits of serverless with enhancements for AI, like
  • GPU Functions and GPU Sandboxes.
  • Automatic model mounting for streamlined self-hosting.
  • An SDK for sandboxing untrusted agent actions.
  • Built-in AI frameworks, with runtime dependency installs for dynamic agents.
  • Low latency and minimal time‑to‑first‑token for fast, reliable start times.
Unlike traditional serverless platforms, Buildfunctions treats compute as a nested primitive. You can deploy top-level apps (CPU or GPU Functions) that instantly spin up hardware-isolated environments (CPU or GPU Sandboxes) for untrusted AI actions, such as running AI-generated code or RL reward functions. Whether you’re building a simple API or a multi-agent system requiring GPU acceleration, Buildfunctions provides the secure, scalable primitives you need.

​
Traditional Serverless Wasn’t Built for Agents

AI agents push beyond what traditional serverless was designed for, including dynamic execution of untrusted actions, long-running stateful workflows, and autonomous behavior that introduces performance, cost, and security challenges. The problem is that agents fail mid-task when they need new libraries, run longer than expected, or pass files between steps, all of which are blocked by rigid runtimes, hard timeouts, and ephemeral filesystems. In worst cases, a bad loop can starve other workloads and trigger autoscaling that turns a single mistake into a massive cloud bill. The gap is that legacy serverless lacks the fundamentals required to self-host AI agents, such as automatic model mounting, GPU access inside sandboxes, and a unified setup. As a result, most teams building agents end up stitching together multiple providers for hosting, sandboxed execution, and inference, each with separate contracts, billing, and operational overhead.

​
Buildfunctions: The Unified Solution for Building Resilient AI Applications 🪐

✔ Predictable performance with full resource isolation ✔ Hard cost controls to prevent runaway spend ✔ Strong security isolation across networks, environments, and credentials ✔ Stateful, multi-step agent workflows that actually complete ✔ The potential for concurrent requests on the same GPU for significant cost savings ✔ One unified platform instead of a fragmented multi-vendor stack

​
Core Concepts

Functions vs Sandboxes:
  • Functions (CPUFunction, GPUFunction): Orchestrate top-level application or agent logic.
  • Sandboxes (CPUSandbox, GPUSandbox): Execute untrusted and dynamic agent actions with full GPU access, automatic model mounting, built-in AI frameworks, runtime dependency installs, and more. They spin up instantly for isolated tasks and can run for up to 24 hours.

​
Architecture Pattern

To simplify your architecture, we recommend this consistent pattern:
  • Top-Level: Use CPUFunction or GPUFunction for deployment on Buildfunctions.
  • Nested: Use CPUSandbox or GPUSandbox when executing work within an existing application (Buildfunctions or anywhere else).

Create a free account

Stop wasting time on servers and environments, and start building functions today!