AnyFrame

AnyFrame is a runtime infrastructure layer for AI agents, built by the team at TinyHQ and accessible at anyfrm.com. It provides isolated microVM sandboxes for agent harnesses like Claude Code, OpenAI Codex, and Cursor, letting each agent run in its own environment that can be paused and resumed without losing state. The product is currently available with a free tier and no credit card required to get started.

What It Is

AnyFrame sits between your agent harness and the underlying compute, wrapping each agent task in a microVM called a "frame." A frame encapsulates the full runtime environment — filesystem, processes, memory, and open file handles — so it can be snapshotted to disk in under one second and resumed exactly where it left off, hours or days later. This makes AnyFrame a purpose-built solution for long-running or asynchronous agent workloads where standard cloud VMs or containers would lose state on pause or restart.

How the Sandbox Model Works

Each frame is an independent microVM, not a container or shared process. When you pause a frame, the entire sandbox state is written to disk. When you resume it, the agent picks up at the next instruction with no re-initialization. The fleet dashboard at anyfrm.com/fleet shows all running, paused, and cold frames in a single view with live CPU, memory, uptime, and the current command in flight — no SSH or cloud console access required. Frames are sized to the task (vCPU and RAM configurable) and billed only for active runtime.

SDK and Programmatic Spawning

AnyFrame ships a Python SDK (uv add anyframe) that lets developers spawn parallel fleets of frames from code. A typical pattern creates one frame per incoming job, attaches an agent session, and lets each run to completion independently. The SDK exposes af.agents.create() and af.sessions.create() calls, making it straightforward to integrate into CI pipelines, eval harnesses, or any batch workload that benefits from parallelism.

Integrations and Connectors

AnyFrame supports a set of pre-built connectors that can be authorized once and toggled per agent. The homepage lists Linear, GitHub, Slack, Notion, Atlassian, Gmail, Figma, Google Calendar, and Google Drive as available connectors. MCP (Model Context Protocol) servers are also supported, giving agents access to custom tool registries alongside the built-in connector set.

Web Chat Widget

The open-source anyframe-web-chat repository on GitHub provides a drop-in chat widget that lets any website give visitors a private AI agent conversation, each running in its own isolated AnyFrame sandbox. A single <script> tag embeds the widget; a lightweight server holds the API key and manages sessions. Each visitor's sandbox is snapshotted between messages so reloads resume the conversation. The widget is unbundled JavaScript and CSS with no build step, and the server is designed for single-replica Railway or Docker deployments.

Current Status

AnyFrame is actively developed by TinyHQ. The anyframe-web-chat repository was created in May 2026 and last updated shortly after. The main product at anyfrm.com is live and accepting signups without a credit card. Community support is available via a Discord server, and the Python SDK is published to PyPI.