Skip to main content

Sandboxes

Your coding agents. Clear limits.

Use familiar coding tools with access your organization defines.

In early access, run Claude Code and Codex on remote machines with chosen repositories, models, tools, and budgets.

Coding agents need access that fits the task.

A developer laptop may hold credentials and projects beyond the current assignment. A separate coding environment can help your team limit the access available to supported agent work.

credentials taken from 1,079 developer machines in one attack
2,349
secrets leaked across 487 organizations by the next one
14,000
of published agent skills scanned were malicious
11.9%
Nx s1ngularity, 2025 · Shai-Hulud, 2025 · ClawHub, 2026 · CVE-2025-59536

The same command, off the developer laptop.

  • Type what you type today

    Start the session from the repository directory. Same terminal, same shortcuts, same full screen.

  • Close the lid, keep the session

    The session lives on the remote machine. Reattach and it is where you left it.

  • Your checkout stays yours

    The agent works on the branch in the sandbox. Your editor keeps its own copy.

Example — fictional team and data. Not a live run or customer result.

Example terminal session

Live
  1. claude
  2. sandbox ready · policy "Engineering default" v7 · egress restricted
  3. Claude CodeReading src/billing/invoice.ts
  4. Claude CodeEdit src/billing/invoice.ts
  5. Blocked by policy: api.unknown-host.dev
  6. Claude CodeRan tests · 41 passed

Ready in 6.2s

One policy decides what a session can reach.

  • Repositories and network

    Choose which repositories and network destinations the governed session may reach. Review requests blocked by those limits.

  • Choose a sign-in mode that fits

    Review the supported provider and developer-subscription options. Credential access, limits, and billing depend on the mode you choose.

  • Tools, skills, and spend

    Choose the supported tools, models, instructions, and budget available to the session. Review changes as the work’s needs change.

Example — fictional team and data. Not a live run or customer result.

Sandbox policy

Engineering default · v7

Repositories
acme-api, acme-web
Network
Restricted · 12 hosts
Credentials
Access set by session mode
MCP servers
Aestus, GitHub
Skills
Org catalog · 9
Models
3 allowed
Budget
8h / developer / week
37

refused this week

Scale the work. Keep the reach small.

Separate coding work from unrelated laptop access. Set the limits for supported sessions and review their activity as usage grows.

  • Limit access before work begins

    Review the repositories, services, and credentials a session needs. A smaller scope can help limit the effect of a mistaken action.

  • A consistent setup across the team

    Give developers a shared starting point for coding tools and access, with less need to set up each environment by hand.

  • Manage idle time and spend

    Set time and spending limits for sessions. Review active usage and storage costs when planning capacity.

Review the session alongside the work.

Use the available terminal recording, tool activity, blocked requests, and cost record to review a coding session. Coverage depends on the supported tools and session mode.

Measure risk and cost from the same run.

Credential access by session mode
Review which credentials are available in the chosen mode and whether that access fits the task.
Requests refused at the boundary
Network and credential calls stopped by policy.
Active hours by developer and repository
Review active session time by developer and repository alongside the applicable usage charges.

Anthropic and OpenAI already ship sandboxes.

Those tools offer useful controls. Aestus adds a remote workspace with organization-set access and a session record. Compare the available modes against your team’s requirements.

Govern one agent session.

Talk to us