Launching today
Sandock
Persistent Docker Workspaces for Coding Agents
18 followers
Persistent Docker Workspaces for Coding Agents
18 followers
Your coding agent needs a real filesystem — not a sandbox that forgets everything. Sandock gives it one: long-running Docker containers with SSD-backed persistent volumes and 100% POSIX compatibility. Run Claude Code, Codex, Anthropic Agent SDK, OpenClaw, Hermes, Pi Coding Agent, or your own agents directly — or embed them into your product. Terminal, browser, preview, git, and MCP workflows are built in, with pricing from $5/mo.
Hey PH! 👋
We were building a coding agent on top of Claude Code. It kept failing — not because the code was wrong, but because the sandbox didn't have a real filesystem. Symlinks broken. File permissions wrong. npm install failing silently.
Most container-based sandboxes have this problem: incomplete POSIX support. And they forget everything when they close.
We tried the alternatives. One had great DX but cost $150+/mo minimum — and still had POSIX issues. Another was powerful but way too complex, clearly built for enterprise teams, not indie builders. Neither gave us persistent, POSIX-complete storage at a reasonable price.
We needed something persistent like a VM, fast like a container, POSIX-complete like a real machine, and actually affordable. Nothing existed. So we built it.
**What makes Sandock different:**
💾 **100% POSIX + Persistent SSD Volumes** — symlinks, permissions, hard links all work. Claude Code, Codex, Anthropic Agent SDK — zero config, no workarounds.
⚡ **Sub-second warm starts, per-second billing** — containers stay alive between calls. $5/mo vs $150+/mo with alternatives.
🔌 **MCP Server Hosting + Simple SDK** — 5 lines of Python or Node.js. Persistent MCP servers. No Docker expertise required.
We've been running our own agents on Sandock for months. It's the infrastructure we wished existed when we started.
Your coding agent deserves a real home. What's your biggest pain point with agent sandboxes right now? 🐳
Sandboxing is becoming one of the essential pieces of infrastructure for AI agents. To keep execution environments safe, most agents need a reliable sandbox to run code, install packages, and interact with files.
What I really like about Sandock is that it focuses on the practical pain points developers actually hit: 100% POSIX support, persistent SSD volumes, and an SDK that looks easy to integrate. Those are exactly the kinds of things that make life easier when building real AI agents instead of just demos.
Excited to see more infrastructure like this emerge for the agent ecosystem. Congrats on the launch! 👍
Sandock feels like an important infrastructure layer for AI coding agents: fast startup, isolated execution, and a familiar developer environment.