AnyFrame
Sandboxes for AI Agents
Last verified:
What is AnyFrame?
AnyFrame is a runtime layer and control plane for AI agents that provides isolated, pausable microVM-based sandboxes. It allows developers to point AI coding agents like Claude Code, Codex, or custom harnesses at any repository and get a fresh sandbox running in seconds. The platform bakes cached sandbox images from a single agent definition that includes the repo, install commands, skills, and MCP connectors.
Key features include a web UI and Python SDK for driving agent sessions on the same channel, fleet visibility showing running/paused/cold frames with CPU/memory/uptime metrics, no-SSH access without cloud consoles, and MCP connector integration (Linear, Sentry, Google) that can be toggled per-agent. The platform supports pausing and resuming agents with state preserved by snapshotting RAM and processes to disk in under a second.
AnyFrame is designed for software engineers, AI/ML developers, indie hackers, and teams building AI agents or coding assistants. Use cases include internal deploy-from-chat agents, scoped coding agents locked to specific repos, per-PR review bots that actually run code, and any scenario requiring isolated sandbox environments for parallel agent execution.
AnyFrame pricing
Pricing model: Freemium
Free trial available without credit card. Users can try for free by bringing a Claude OAuth token or Codex API key. Paid plan pricing requires contacting the team (to-be-determined). The product is currently in PreOrder status with pricing details to be confirmed on the official website.
AnyFrame pros
- Boots sandbox sessions in seconds
- Define an agent once and reuse it
- Cached sandbox images for faster startup
- Web UI and Python SDK on same channel
- Supports Claude Code, Codex, and custom harnesses
- Fleet dashboard shows all frames in one view
- Monitor CPU, memory, and uptime visually
- No SSH or cloud consoles needed
- Pause and resume with state preserved
- Snapshots RAM and processes to disk in under a second
- MCP connectors (Linear, Sentry, Google) plug in once
- Toggle MCP connectors per-agent
- Free trial without credit card required
- Model-agnostic platform for any AI agent
- Enables parallel fleet of isolated sandboxes
AnyFrame cons
- Early version still in development
- Pricing details not publicly disclosed
- Requires Claude OAuth token or Codex API key
- PreOrder status indicates unreleased pricing
- Limited documentation available
- New product with small user base
- No native mobile app
- Python SDK only (no other language SDKs)
Frequently asked questions about AnyFrame
What is AnyFrame used for?
AnyFrame is used to run, manage, and scale stateful AI agents in isolated, pausable sandboxes. It is ideal for development, testing, and production deployment of autonomous AI workflows, allowing teams to avoid hand-rolling the same sandbox plumbing for every agent.
How does AnyFrame's pausing technology work?
AnyFrame uses microVM technology to snapshot an entire agent's environment—including RAM and processes—to disk in under a second. This allows you to stop paying for compute while preserving the exact state for later resumption.
Can I run different AI models like Claude and Codex on AnyFrame?
Yes, AnyFrame is model-agnostic. It can host any AI agent harness, including Claude Code, OpenAI Codex, Cursor, and custom agents, each in its own isolated sandbox.
How does AnyFrame handle multiple concurrent agents?
AnyFrame's Python SDK and fleet dashboard allow you to spawn and manage a parallel fleet of sandboxes, assigning one microVM per task or agent for isolated, scalable execution.
Is there a free trial for AnyFrame?
Yes, AnyFrame offers a free trial to spin up your first frame in seconds without requiring a credit card. You need to bring a Claude OAuth token or Codex API key to get started.
What MCP connectors does AnyFrame support?
AnyFrame supports MCP connectors including Linear, Sentry, Google, and others. You plug in MCP connectors once and can toggle them per-agent as needed.
How do I authenticate with AnyFrame's Python SDK?
The SDK authenticates with a personal API token with the prefix afm_. You sign in at anyfrm.com, open Dashboard → Settings → API keys, and click Create key.
What?'s the difference between running, paused, and cold frames?
Running frames are actively executing, paused frames have their state snapshot to disk (saving compute costs), and cold frames are not currently active. The dashboard shows all three states in one view with CPU, memory, and uptime metrics.
Can I build a per-PR review bot with AnyFrame?
Yes, one key use case is creating a per-PR review bot that actually runs your code rather than just reading it. You define the agent with the repo and install commands, and it runs in an isolated sandbox.
How do I get access to AnyFrame?
You can visit anyfrm.com to sign up for the free trial. Bring a Claude OAuth token or Codex API key and you're in. For paid plans or more information, join their Discord at discord.gg/UpkEW6JjpU or check the Product Hunt page.