llms.txt
# AnyFrame
> Runtime layer for AI agents. Spawn sandboxes at the speed of natural language. Pause, resume, and snapshot the full state of an agent's workspace — filesystem, processes, and memory — and pick up later without losing context.
AnyFrame is a sandbox platform built for long-running AI agents. Every frame is its own isolated sandbox with a complete developer environment (Node.js, Python, Go) and one of the supported agent harnesses pre-installed. State persists between sessions.
## Quick Start
- [Changelog](https://anyframe.dev/changelog)
- [GitHub](https://github.com/tinyhq)
- [Discord](https://discord.gg/nn9hmgemdY)
- Request early access: https://anyframe.dev/#waitlist
## Core Concepts
- **Frame**: a sandbox with a filesystem, running processes, and memory.
- **Snapshot**: the full state of a frame serialized to disk. Resuming a snapshot restores the VM exactly where it stopped.
- **Harness**: the AI coding agent installed in a frame (Claude Code, Cursor, Codex CLI).
## Core SDK (TypeScript / Python)
- Create a frame: `await af.frames.create({ harness: 'claude-code' })`
- Exec a command: `await frame.exec('npm test')`
- Pause a frame: `await frame.pause()` // snapshots state, releases compute
- Resume a frame: `await frame.resume()` // restores state from snapshot
- Fork a frame: `await frame.fork()` // branches a paused frame
- List frames: `await af.frames.list()`
## Supported Harnesses
- **Claude Code** (Anthropic) — agentic coding via the Claude API
- **Cursor** (Anysphere) — Cursor's agent in headless mode
- **Codex CLI** (OpenAI) — OpenAI's coding CLI
Each frame ships with Node.js, Python, and Go pre-installed and the chosen harness ready to run. Swap harnesses on the same frame to hand work between agents without losing context.
## Key Features
- **Pause and resume in under a second**: full filesystem, process, and memory snapshot.
- **State persists between sessions**: a paused frame can sit for days; resume picks up the next instruction.
- **Fleet view**: see every cold, paused, and running frame in your account in one place — no SSH gymnastics.
- **Per-job isolation**: each agent task lands on its own sandbox with the resources it needs, runs to completion, and snapshots away when done.
- **SDK + REST API**: TypeScript and Python wrappers, plus raw HTTP for everything else.
## Use Cases
- Long-running agent sessions that span multiple days or working sessions.
- Parallel agent fleets — spawn one frame per task, run independently, snapshot results.
- Real-time production debugging without standing up new infrastructure.
- Branching: fork a paused frame to explore an alternative path while keeping the original intact.
- Hand-offs: pause a Claude Code frame, hand the snapshot to Cursor, resume.
## Pricing
Currently in invite-only early access. Pricing tiers will be announced when we move to general availability. To request access: https://anyframe.dev/#waitlist
## Privacy and Security
- Workspace contents (your code, files, processes inside a frame) are encrypted at rest and never read by us, never used to train models.
- We log frame metadata (resource usage, timestamps, request IDs) for billing and incident response.
- Subprocessors: Ubicloud (compute/DB), GitHub (OAuth), Cloudflare (edge), Google Apps Script (waitlist intake).
- Full details: https://anyframe.dev/privacy
## Acceptable Use
- No cryptocurrency mining, no spam infrastructure, no unauthorized scanning or DoS.
- No attempts to escape frame isolation or access other customers' data.
- No reverse engineering of the platform itself.
- Full Acceptable Use Policy: https://anyframe.dev/terms
## Company
- Website: https://anyframe.dev
- GitHub: https://github.com/tinyhq
- Discord: https://discord.gg/nn9hmgemdY
- Changelog: https://anyframe.dev/changelog
- Privacy Policy: https://anyframe.dev/privacy
- Terms of Service: https://anyframe.dev/terms
- Contact: founders@anyframe.dev
