Podflare

Podflare provides hardware‑isolated microVM sandboxes that give AI agents a dedicated Linux kernel, root access, outbound internet, and a writable filesystem, enabling secure, low‑latency code execution.

Updated 2 months ago

Funding

Funding not disclosed

Funding rounds are not available yet.

Founders

Founder details are not available yet.

Product

Problem

AI agents that need to execute generated code require a secure, low‑latency environment with full Linux capabilities, persistent state, and the ability to install arbitrary packages. Existing container or serverless solutions either share the host kernel, introduce high cold‑start latency, or lack features such as sandbox forking for tree‑search workflows.

Solution

Podflare delivers hardware‑isolated microVM sandboxes that provide each agent with a dedicated Linux kernel, root access, outbound internet, and a writable filesystem. The platform offers sub‑millisecond hot‑execution latency for successive `run_code` calls, preserving a persistent Python REPL so variables, imports, and files survive across tool invocations. A copy‑on‑write snapshot primitive (`fork(n)`) enables rapid cloning of a running sandbox in ~80 ms, supporting efficient tree‑search and multi‑branch agent strategies. Podflare integrates natively with major AI SDKs—including OpenAI Agents, Anthropic code_execution, Vercel AI SDK, and MCP—through simple adapters, allowing agents to install any pip, npm, or apt package without allow‑list restrictions. The service scales across configurable tiers (1 GB to 16 GB RAM, up to 64 GB disk) and regions, with predictable pricing and optional egress‑disabled sandboxes for air‑gapped workloads.

Target Audience

Podflare targets developers building LLM‑driven agents, autonomous AI assistants, and tool‑use pipelines that require secure, low‑latency code execution, as well as enterprises needing air‑gapped sandboxes for handling sensitive data.

Features

Hardware‑isolated microVMs with a dedicated Linux kernel, root privileges, and full internet access
Persistent Python REPL that retains variables, imports, and open files across `run_code` calls
Ultra‑fast hot‑execution (~60 ms) and sub‑millisecond latency for repeated tool calls
Copy‑on‑write sandbox snapshot (`fork(n)`) completing in ~80 ms to spawn multiple divergent instances
Unlimited package installation via pip, npm, apt, or git without allow‑list constraints
Writable filesystem (4 GB default, up to 64 GB) with upload/download APIs for data movement
Native adapters for OpenAI Agents SDK, Anthropic code_execution, Vercel AI SDK, and MCP enabling drop‑in integration
Tiered resource configurations (RAM, vCPU, disk) and region selection with per‑sandbox concurrency limits