Introducing Context Hosting

The thesis is simple: 1x monetary value → 100x agents → 100,000x sandboxes.

As AI coding agents move from demos to production, sandbox infrastructure becomes the critical bottleneck. Every agent needs a real execution environment — filesystem access, network, tools — but you can’t give them the keys to production.

The Problem

Running a single coding agent on your laptop is easy. Running 100 agents across a team, each working on different repos, each needing isolated environments with secrets management and state persistence — that’s an infrastructure problem.

Existing solutions either:

• Restrict agents too much — sandboxes without real tool access are useless
• Don’t scale — running Docker containers locally breaks at 10+ agents
• Lack observability — you can’t see what 50 agents are doing in parallel

Our Approach

Context Hosting gives every agent a full Kubernetes pod with:

• Real access — filesystem, network, tools, all isolated per-sandbox
• Multi-agent support — Claude Code, OpenAI Codex, Amazon Bedrock
• State management — snapshot, restore, fork at any point
• Unified interface — same tmux session from browser or CLI
• Boring tech — Kubernetes, Postgres, Linux. No magic.

# Spin up an agent in seconds
ctx run my-sandbox --agent codex

# Attach to it
ctx attach my-sandbox --session agent

What’s Next

We’re building in the open. The full source is on GitHub.

Try it out, file issues, and help us build the infrastructure layer for the agent era.