Heads up This site is currently under heavy development.
← all tools
◆ AI/LLM Security

Docker Sandboxes

v0.39.0 commercial

Docker Sandboxes isolates containerized applications in secure, lightweight environments for testing and development without affecting the host system.

Summary

Docker Sandboxes is an open-source command-line tool that runs AI coding agents inside isolated microVM sandboxes, each with its own Docker daemon, filesystem, and network so agents can build containers, install packages, and modify files without touching the host. The `sbx` CLI itself is free, including for commercial use; centrally managing network and filesystem policy across an organization's developers from the Docker Admin Console requires a separate paid subscription. It installs via Homebrew, WinGet, or APT/RPM packages and works with Claude Code, Codex, Gemini CLI, OpenCode, and other agents, making it a fit for developers and platform teams who need to run autonomous coding agents without exposing their machines to arbitrary file or network access. It sits in the runtime-security and sandboxing space, layering egress policy, secrets management, and audit logging around agent execution rather than just isolating processes.

Docker Sandboxes isolates containerized applications in secure, lightweight environments for testing and development without affecting the host system.

What Docker Sandboxes answers

What stops an agent from reaching the internet or exfiltrating data through the back door?

network policy gates DNS resolution and blocks outgoing ICMP by default, closing lookup- and ping-based exfiltration channels even if a domain is otherwise denied

What operating systems and hardware can actually run this?

packages ship for macOS, Windows, Ubuntu, Rocky Linux, and Linux ARM64, though Linux hosts need kvm group membership since sandboxes are microVMs rather than plain containers

Can I keep using the models and secrets I already have configured?

built-in providers like OpenRouter get automatic bearer-token injection, custom secrets support wildcard host patterns and Basic-auth substitution, and Linux falls back to an encrypted on-disk store when no keychain is present

Do I have to trust every kit I find, or can I lock that down?

kit installs are restricted to an allowlist of sources, defaulting to Docker Hub only, so pulling from another registry or a Git URL fails until an admin explicitly permits it

Can I prove to auditors what the agent was allowed to touch?

policy rules can be listed per sandbox with inactive governed rules revealed on demand, and structured audit logs of every policy decision can be forwarded to a SIEM, though that logging requires the paid governance subscription

Will multiple sandboxes for the same project step on each other?

sandboxes can be named independently of the working directory, so several run side by side in one workspace and can be re-attached from any directory by name

all 21 features, with the evidence for each →

Features

21 capabilities across 5 areas

Built from everything we hold on Docker Sandboxes — every release we have summarised, its product documentation and how that documentation has changed, its README, its command-line surface and API, and runs we performed ourselves. Dates are when we first saw a capability, not when the vendor introduced it.

Capability area
Sandbox Lifecycle and Access 8 capabilities Covers the core operations for creating, running, inspecting, and tearing down sandboxes. Gives operators and agents consistent, scriptable control over sandbox state from creation through deletion.
SSH endpoint for sandbox access shipped Exposes a native SSH endpoint so agents, operators, and editors can open a shell directly into a running sandbox. 2 releases · 11 other sources · first seen Jun 2026

release

  • Adds SSH access to sandboxes (experimental): run sbx setup ssh once to register a managed *.sbx SSH config entry, then connect to any sandbox with ssh <name>.sbx from OpenSSH clients or remote-dev tools like VS Code and Cursor. v0.37.0 · Jul 2026 · source · release history
  • New experimental native SSH endpoint in sandboxd: connect with ssh <sandbox-name>@127.0.0.1 -p 2222 (publickey auth, connect-to-create, interactive shell and exec). v0.34.0 · Jun 2026 · source · release history

docs 7 total

  • Manuals Get started Guides Reference Connect ChatGPT to a sandbox Table of contents Availability: GA Requires: Docker Sandboxes 0.37.0 or later Connect the ChatGPT desktop app to a sandbox over SSH so Codex works inside the isolated environment instead of on your host. docs-2c0c7a4669f9 · Jul 2026 · source
  • Manuals Get started Guides Reference Connect Cursor to a sandbox Table of contents Availability: GA Requires: Docker Sandboxes 0.37.0 or later Cursor is built on VS Code, so it connects to a sandbox the same way, using Remote - SSH. docs-dd8686b94f33 · Jul 2026 · source
  • Manuals Get started Guides Reference Connect Claude Desktop to a sandbox Table of contents Availability: GA Requires: Docker Sandboxes 0.37.0 or later Claude Desktop can run Claude Code on a remote machine over SSH. Point it at a sandbox so the agent works inside the isolated environment instead of on your host. docs-abd27a552b81 · Jul 2026 · source
  • Manuals Get started Guides Reference Editor and app integrations Table of contents Availability: GA Requires: Docker Sandboxes 0.37.0 or later You can connect an external editor or desktop app to a running sandbox over SSH. docs-3ad7676b2230 · Jul 2026 · source
  • Manuals Get started Guides Reference Connect VS Code to a sandbox Table of contents Availability: GA Requires: Docker Sandboxes 0.37.0 or later Use the Remote - SSH extension to open a VS Code window that runs inside a sandbox. docs-d4ae7567788b · Jul 2026 · source
  • Integrations — connect editors and apps like VS Code and Cursor to a sandbox over SSH docs-571863b279ad · Jul 2026 · source

example

  • Set up SSH access once so every sandbox becomes reachable by name from any OpenSSH-compatible client or remote-dev tool. sbx setup ssh v0.37.0 · Jul 2026 · source
  • Connect to a running sandbox for an interactive shell or one-shot command without leaving your terminal. ssh my-sandbox.sbx v0.37.0 · Jul 2026 · source
  • Enable the experimental SSH endpoint so agents or operators can shell directly into a running sandbox without a separate proxy. sbx settings set feature.ssh true v0.34.0 · Jun 2026 · source
  • Connect an interactive shell to a named sandbox over the native SSH endpoint once feature.ssh is enabled. ssh <sandbox-name>@127.0.0.1 -p 2222 v0.34.0 · Jun 2026 · source
Named sandbox management shipped Lets users create, reconnect to, and list sandboxes by a stable name regardless of where they run the command. 6 releases · 2 other sources · first seen Jun 2026

release

  • Re-attaching to a sandbox created with a custom --kit agent now works via sbx run --name <name> without re-passing --kit. v0.35.0 · Jul 2026 · source · release history
  • sbx ls --json now reports a stable per-sandbox id field for scripting and automation. v0.33.0 · Jun 2026 · source · release history
  • The sbx run <name> positional argument form for re-attaching to an existing sandbox now prints a deprecation warning; the positional argument for sbx run is expected to be an agent (e.g. claude, codex), not a sandbox name, and the old positional-name form will be removed in a future release. v0.33.0 · Jun 2026 · source · release history
  • sbx run --name no longer auto-creates numbered sibling sandboxes when a same-named sandbox exists. v0.33.0 · Jun 2026 · source · release history
  • sbx create now errors when run without arguments instead of proceeding. v0.33.0 · Jun 2026 · source · release history
  • New --name flag for sbx run identifies a sandbox independently of the working directory, enabling multiple named sandboxes in the same workspace and re-attachment from any directory. v0.33.0 · Jun 2026 · source · release history

example

  • Re-attach to a named sandbox from a different directory without specifying an agent, useful when returning to a long-running session. sbx run --name feature-a v0.33.0 · Jun 2026 · source
  • Run two independent Claude Code sandboxes in the same workspace — useful when parallelising agent tasks without them colliding. sbx run --name feature-a claude && sbx run --name feature-b claude v0.33.0 · Jun 2026 · source
Sandbox filesystem operations shipped Lets users copy files in and out of a sandbox and run commands inside it with consistent working-directory behaviour and faster I/O. 6 releases · first seen Jun 2026

release

  • Adds DOCKER_SANDBOXES_CLONED_WORKSPACE_SIZE environment variable to configure the size of the cloned workspace volume. v0.38.0 · Aug 2026 · source · release history
  • Extends sbx mount <sandbox> <host> so omitting a container target bind-mounts the host path at the same path inside the sandbox; :ro mounts are enforced read-only at the virtio-fs host edge. v0.38.0 · Aug 2026 · source · release history
  • Enables virtiofs caching by default on all operating systems for faster filesystem performance (opt out with DOCKER_SANDBOXES_ENABLE_VIRTIOFS_CACHE=0). v0.35.0 · Jul 2026 · source · release history
  • sbx exec now uses the same working directory as sbx run. v0.33.0 · Jun 2026 · source · release history
  • sbx cp -L now follows symlinks in the source path for sandbox-to-host copies. v0.33.0 · Jun 2026 · source · release history
  • virtiofs cache is now enabled by default on macOS and Linux, improving filesystem performance. v0.33.0 · Jun 2026 · source · release history
Linux arm64 and headless platform support shipped Supports Linux arm64 builds and headless server deployments with an encrypted on-disk keychain when a desktop session is unavailable. 3 releases · first seen Jun 2026

release

  • Homebrew installs now use a stapled .dmg artifact instead of a .tar.gz archive, improving Gatekeeper compatibility on macOS. v0.38.0 · Aug 2026 · source · release history
  • New keychain fallback to encrypted on-disk store on Linux when dbus-launch is unavailable, enabling headless/server deployments. v0.33.0 · Jun 2026 · source · release history
  • Linux arm64 build packages are now produced, extending platform support. v0.33.0 · Jun 2026 · source · release history
Sandbox port publishing shipped Publishes sandbox ports to the host at creation time and recovers automatically from port conflicts on daemon restart. 2 releases · 1 other source · first seen Jun 2026

release

  • Adds -p/--publish flag to sbx create and sbx run for publishing sandbox ports at creation time. v0.37.0 · Jul 2026 · source · release history
  • Restores published sandbox ports on restart, with automatic recovery from host-port conflicts by selecting a new host port. v0.34.0 · Jun 2026 · source · release history

example

  • Start a sandbox with a published port and without shared skills, e.g. for an isolated web-service test. sbx run -p 8080:8080 --no-share-skills my-image v0.37.0 · Jul 2026 · source
Daemon lifecycle management shipped Provides CLI subcommands to start, stop, check status, and adjust logging for the background sandbox daemon. 2 releases · first seen Jul 2026

release

  • Adds sbx daemon restart to stop and restart the sandboxd daemon in the background. v0.38.0 · Aug 2026 · source · release history
  • New sbx daemon command with start, stop, status, and log-level subcommands for daemon lifecycle management. v0.35.0 · Jul 2026 · source · release history
Sandbox inspection shipped Shows a sandbox's installed kits, injected secrets, published ports, and runtime details in a single command. 2 releases · first seen Jul 2026

release

  • sbx inspect now displays custom secrets configured for a sandbox. v0.38.0 · Aug 2026 · source · release history
  • sbx inspect now surfaces the sandbox's kits, injected secrets, and sandbox information. v0.35.0 · Jul 2026 · source · release history
Safe sandbox deletion shipped Prevents accidental deletion of a running sandbox by requiring an explicit force flag before teardown. 2 releases · first seen Jul 2026

release

  • sbx rm now won't delete an active session unless --force is passed. v0.35.0 · Jul 2026 · source · release history
  • sbx rm now requires --force to delete an active session, preventing accidental teardown. v0.35.0 · Jul 2026 · source · release history
Network Control and Egress Security 5 capabilities Defines and enforces what network traffic sandboxes are allowed to send or receive. Combines policy tiers, active blocking of covert channels, upstream proxy routing, and TLS inspection into a layered egress control model.
Sandbox network policy shipped Lets operators define, inspect, test, and list egress network policies for sandboxes, with preset tiers and per-request policy checks before any outbound connection is made. 8 releases · 3 other sources · first seen Jun 2026

release 8 total

  • Adds --deny-network HOST flag to sbx run and sbx create for per-sandbox egress deny rules recorded at creation time. v0.38.0 · Aug 2026 · source · release history
  • Blocked HTTPS proxy connections now appear in sbx policy log even when the client aborts the TLS handshake. v0.38.0 · Aug 2026 · source · release history
  • New sbx policy check network command to test whether the current policy would allow a network access request before executing. v0.35.0 · Jul 2026 · source · release history
  • Balanced network preset now allows VS Code domains, Azure Blob Storage (*.blob.core.windows.net), and dhi.io over HTTP. v0.35.0 · Jul 2026 · source · release history
  • Revamped sbx policy ls with --wide, --source, and --decision filters for concise policy listing. v0.35.0 · Jul 2026 · source · release history
  • New sbx policy inspect command to examine policy details. v0.35.0 · Jul 2026 · source · release history

docs

  • Configure a support message Admins can add an optional support message that appears after the policy denial details when a sandbox action is blocked by organization governance. Use it to point members to an internal support channel, ticket queue, or security contact. docs-941da611c5e3 · Jul 2026 · source

example

  • Block a sandbox from reaching a specific host at creation time, e.g. to prevent exfiltration to an external service. sbx run --deny-network api.example.com claude v0.38.0 · Aug 2026 · source
  • Before running an agent task, verify that the current policy permits a specific outbound connection — catch misconfigurations without launching the full sandbox. sbx policy check network --host api.example.com --port 443 v0.35.0 · Jul 2026 · source
Upstream proxy support for sandbox egress shipped Routes sandbox outbound traffic through a corporate HTTP or SOCKS5 upstream proxy, with optional credentials and exclusion lists. 4 releases · 2 other sources · first seen Jul 2026

release

  • Adds proxy.integratedAuth setting to enable NTLM/Kerberos/Negotiate proxy authentication on Windows. v0.38.0 · Aug 2026 · source · release history
  • Adds settings-driven upstream-proxy configuration via proxy, proxy.sandbox, proxy.daemon, and matching no_proxy keys, defaulting to the host OS system proxy and proxying daemon traffic (image pulls, telemetry) as well. v0.38.0 · Aug 2026 · source · release history
  • New DOCKER_SANDBOXES_PROXY=system env var routes sandbox egress through the host OS proxy (macOS/Windows), including PAC auto-config URLs. v0.37.0 · Jul 2026 · source · release history
  • Adds SOCKS5 upstream-proxy transport (socks5:// / socks5h://) for sandbox egress via DOCKER_SANDBOXES_PROXY, HTTP_PROXY, or HTTPS_PROXY, with optional auth and DOCKER_SANDBOXES_NO_PROXY exclusions. v0.35.0 · Jul 2026 · source · release history

example

  • Configure a corporate HTTPS proxy for all sandbox and daemon traffic with NTLM authentication on Windows. sbx settings set proxy http://proxy.corp.example.com:8080 sbx settings set proxy.integratedAuth true v0.38.0 · Aug 2026 · source
  • Route all sandbox egress through a corporate SOCKS5 proxy (with credentials) so sandbox traffic flows through your org's inspection infrastructure. DOCKER_SANDBOXES_PROXY=socks5://proxyuser:[email protected]:1080 DOCKER_SANDBOXES_NO_PROXY=localhost,127.0.0.1 sbx run v0.35.0 · Jul 2026 · source
Audit and execution-outcome logging shipped Emits structured audit records for network egress and filesystem mount events, capturing success, latency, and error class for every event. 1 release · 2 other sources · first seen Jul 2026

release

  • Audit now emits execution-outcome records for network egress (per allowed connection) and filesystem mounts (per allowed path), including success, latency, and error class. v0.37.0 · Jul 2026 · source · release history

docs

  • AI Governance Audit Logs Subscription: AI Governance Requires: Docker Sandboxes 0.35.0 or later AI Governance Audit Logs record Docker AI Governance activity for your organization. Each record captures the principal, action, target, decision, and time for a governance event. Records contain metadata only. docs-c15e55dc633d · Aug 2026 · source
  • Manuals Get started Guides Reference Local audit logs Table of contents The sandbox daemon writes local audit records as JSON Lines ( .jsonl ) files. Local audit logs stay on the host that produced them and can be collected by your own log shipper. docs-28787d551c65 · Aug 2026 · source
MITM TLS proxy shipped Intercepts and decrypts TLS traffic from sandboxes via a built-in proxy, issuing generated certificates so encrypted sessions can be inspected. 3 releases · first seen Jun 2026

release

  • Adds tls.allowNegativeSerial setting for TLS-inspecting proxies that issue certificates with negative serial numbers. v0.38.0 · Aug 2026 · source · release history
  • Faster sandbox startup via optimized TLS-proxy CA installation (merging into trust bundle instead of running update-ca-certificates), saving several hundred milliseconds. v0.35.0 · Jul 2026 · source · release history
  • MITM proxy now publishes a CRL and embeds a CRL distribution point in generated certificates, supporting revocation-strict TLS clients (e.g., .NET). v0.33.0 · Jun 2026 · source · release history
Sandbox network isolation and egress control shipped Blocks unauthorised outbound channels — including ICMP and DNS-based exfiltration — and gates all sandbox egress on the active network policy. 2 releases · first seen Jun 2026

release

  • Outgoing ICMP from sandboxes is now blocked persistently across daemon restarts (CVE-2026-12539). v0.33.0 · Jun 2026 · source · release history
  • Sandbox DNS resolution is now gated on network policy, closing a DNS-based data-exfiltration channel (CVE-2026-12039); loopback names remain exempt. v0.33.0 · Jun 2026 · source · release history
Secrets, Credentials, and Environment Configuration 2 capabilities Handles how sensitive values and pre-packaged tooling reach the inside of a sandbox. Secrets and OAuth credentials are stored securely and scoped to specific sandboxes or host patterns, while kits bundle tooling, rules, and credentials into reusable units.
Sandbox secret and credential management shipped Stores and scopes secrets and OAuth credentials to individual sandboxes or wildcard host patterns, backed by a keychain-protected store. 8 releases · 3 other sources · first seen Jun 2026

release 8 total

  • Service and custom secrets are now global by default, with --sandbox for sandbox-scoped secrets. v0.38.0 · Aug 2026 · source · release history
  • Service and custom secrets default to global scope; legacy positional and --global forms are deprecated with warnings. v0.38.0 · Aug 2026 · source · release history
  • Host env vars no longer auto-inject at runtime — use sbx secret import to migrate existing env-var-based secrets. v0.35.0 · Jul 2026 · source · release history
  • New sbx secret import imports credential env vars into the keychain; sbx secret ls now flags env-only and OAuth-shadowed entries. Host env vars no longer auto-inject at runtime — migration required. v0.35.0 · Jul 2026 · source · release history
  • Kits can inject the user's Docker login token into requests to docker.com hosts via a credential with service sbx-login. v0.35.0 · Jul 2026 · source · release history
  • Droid OAuth credentials are now proxy-managed, keeping real tokens on the host and out of the sandbox. v0.35.0 · Jul 2026 · source · release history

docs

  • OpenCode Zen API keys OpenCode Zen API keys aren't part of the built-in OpenCode credentials that sbx secret set supports. docs-932fdee6aef8 · Aug 2026 · source

example

  • Migrate host env-var credentials (e.g. API keys) into the keychain so they are explicitly managed rather than auto-injected at runtime. sbx secret import v0.35.0 · Jul 2026 · source
  • Grant a custom secret to all subdomains of an internal service using wildcard host patterns. sbx secret set-custom --host '*.internal.corp' --host '**.api.internal.corp' MY_API_TOKEN v0.33.0 · Jun 2026 · source
Kit management shipped Lets operators install, update, and compose kits that extend a sandbox with pre-packaged tooling, network rules, and credentials from an allowlisted set of sources. 4 releases · 1 other source · first seen Jun 2026

release

  • sbx kit add now recreates the sandbox container with the full kit set (preserving state) and applies the kit's network allow/deny rules and composed policy. v0.35.0 · Jul 2026 · source · release history
  • Adds OCI v2 kit artifact streaming that decompresses layers once to a cache directory and uses seek-based random access, reducing memory pressure during kit reads. v0.34.0 · Jun 2026 · source · release history
  • Kit installs are restricted to an allowlist of sources defaulting to Docker Hub (docker.io/) only; installing a kit from another registry or a Git URL fails until you add its prefix with sbx settings set kit.allowedSources. v0.34.0 · Jun 2026 · source · release history
  • Restricts kit installs to an allowlist of sources (default: Docker Hub only); installing from other registries or Git URLs now fails unless explicitly permitted — breaking change. v0.34.0 · Jun 2026 · source · release history

example

  • Allow kit installs from a private registry or Git source in addition to Docker Hub, to avoid the new allowlist breaking change. sbx settings set kit.allowedSources ghcr.io/myorg/ v0.34.0 · Jun 2026 · source
AI Agent Integration 5 capabilities Makes it straightforward to run AI coding agents inside sandboxes with their required services and models working out of the box. Covers OAuth flows, service connectors, skill sharing, MCP server registration, and local model routing.
Agent skills sharing across sandboxes shipped Discovers agent skills on the host and imports them into a persistent store that new sandboxes automatically inherit, with a dry-run preview before committing. 2 releases · 3 other sources · first seen Jul 2026

release

  • Adds sbx skills import to discover and import host agent skills into a persistent store shared across sandboxes, with --dry-run to preview and --force to overwrite existing skills. v0.37.0 · Jul 2026 · source · release history
  • New --no-share-skills flag on sbx run / sbx create lets individual sandboxes opt out of the shared skills store. v0.37.0 · Jul 2026 · source · release history

docs

  • Share agent skills Shared agent skills make skills from supported agents on your host available inside your sandboxes. Importing copies the skills into a persistent store that survives sandbox deletion and is shared by default with new sandboxes that run a supported agent. Shared agent skills are experimental. docs-fce0b00d7ac0 · Jul 2026 · source

example

  • Import host agent skills into the shared store (preview first, then apply) so new sandboxes automatically inherit them. sbx skills import --dry-run && sbx skills import --force v0.37.0 · Jul 2026 · source
  • Start a sandbox with a published port and without shared skills, e.g. for an isolated web-service test. sbx run -p 8080:8080 --no-share-skills my-image v0.37.0 · Jul 2026 · source
MCP server management shipped Lets operators register, list, and remove MCP servers for use inside sandboxes, with live index refresh so new servers are discoverable without recreating the sandbox. 4 releases · first seen Aug 2026

Shared capability: MCP server

release

  • Makes dynamic MCP tools mcp-find, mcp-add, and mcp-config-set available in local sandbox mode. v0.38.0 · Aug 2026 · source · release history
  • Adds sbx mcp add with --dir and OAuth override flags; registering a server immediately refreshes the mcp-find/mcp-add index in every running sandbox so new servers are discoverable without recreating sandboxes. v0.38.0 · Aug 2026 · source · release history
  • The sbx mcp bundle command has been removed; single-server management via sbx mcp add/ls/rm is the only supported flow. v0.38.0 · Aug 2026 · source · release history
  • Adds sbx mcp as a first-class, no-longer-hidden subcommand for MCP server management. v0.38.0 · Aug 2026 · source · release history
Agent connector support shipped Allows recognised AI agents — including Claude Code, Cursor, Droid, and ChatGPT — to run inside sandboxes with their OAuth flows and service connectors working without manual policy overrides. 3 releases · first seen Jun 2026

release

  • New experimental sbx setup command imports agent credentials from environment variables for smoother first-time onboarding. v0.34.0 · Jun 2026 · source · release history
  • Claude connectors (Slack, Gmail, Notion, Atlassian, etc.) now work inside sandboxed Claude Code without manual policy overrides. v0.33.0 · Jun 2026 · source · release history
  • Adds Cursor OAuth support as a recognized agent. v0.33.0 · Jun 2026 · source · release history
Local model routing for agent sandboxes shipped Routes a supported AI agent away from its cloud API to a local model served via llmman or Ollama, enabled through an experimental feature flag. 1 release · 1 other source · first seen Aug 2026

release

  • Adds sbx run --model <name> claude to serve a local GGUF model via llmman, or an already-running Ollama install using an ollama/ prefix, routing Claude Code away from api.anthropic.com; enable with sbx settings set platform.allowExperimentalFeatures true and sbx settings set feature.model true. v0.38.0 · Aug 2026 · source · release history

example

  • Route Claude Code at a local GGUF model instead of api.anthropic.com — useful for air-gapped or cost-sensitive environments. sbx settings set platform.allowExperimentalFeatures true sbx settings set feature.model true sbx run --model llama3.2 claude v0.38.0 · Aug 2026 · source
Claude Vertex multi-region endpoint support shipped Allows Claude Vertex sandboxes to target US or EU multi-region endpoints rather than a single fixed endpoint. 1 release · first seen Aug 2026

release

  • Claude Vertex sandboxes now support us and eu multi-region endpoints. v0.38.0 · Aug 2026 · source · release history
Hardware and Platform Acceleration 1 capability Extends sandboxes beyond standard CPU execution to take advantage of host hardware. Currently covers NVIDIA GPU passthrough for workloads that require GPU access inside the sandbox.
GPU passthrough for sandboxes shipped Enables NVIDIA GPU passthrough via VFIO on Linux so workloads inside a sandbox can access the host GPU directly. 1 release · first seen Aug 2026

release

  • Adds sbx run --gpu (experimental) for NVIDIA VFIO GPU passthrough on Linux; enable with sbx settings set feature.sandbox-gpu true. v0.38.0 · Aug 2026 · source · release history
Capability
Evidence

Lines in monospace are the tool's own words — help text parsed from its source, or an endpoint from its API document. Everything else is our summary of a dated release or documentation change, linked back to the source it came from.

Release history

  1. v0.39.0 Aug 19, 2026 · issue 005

    Docker Sandboxes v0.39.0 adds declarative .sbxenv.yaml environments, sbx env lifecycle commands, USB passthrough, bulk sbx prune, and kit signing via cosign.

    └──▷ GET THIS VERSION
    $ git clone --branch v0.39.0 https://github.com/docker/sbx-releases.git
    # already have the repo? check out this version:
    $ git checkout v0.39.0
    └──▷ TRY IT
    Commit a reproducible sandbox definition with your project so every contributor launches the same environment with one command.
    $ sbx env run
    Define a complete sandbox environment — agent, kits, env vars, secrets, ports, and resource limits — in a project-committed file.
    .sbxenv.yaml
    # .sbxenv.yaml
    agent: claude
    workspace: .
    kits:
      - name: my-kit
    env:
      - NODE_ENV=development
    secrets:
      - MY_API_KEY
    ports:
      - 8080
    Passthrough a specific USB device into a sandbox for hardware-in-the-loop testing on Linux x86_64/ARM64.
    $ DOCKER_SANDBOXES_FEATURE_SANDBOX_USB=1 sbx create --usb /dev/bus/usb/001/002 my-sandbox
    • New sbx env run command provisions a sandbox from .sbxenv.yaml and opens an interactive session; sbx env create, sbx env exec, and sbx env rm manage the full environment lifecycle.
    • New .sbxenv.yaml declarative environment file defines agent, workspace, kits, environment variables, secrets, registry credentials, ports, and resource limits — committable with a project for reproducible contributor environments.
    • Multiple .sbxenv.yaml files can be combined for shared configuration and local overrides, with host environment variable references for machine-specific paths and credentials.
    • New experimental --usb flag on sbx create (enabled via DOCKER_SANDBOXES_FEATURE_SANDBOX_USB environment variable) re-attaches specified USB devices into the sandbox via usbfs on Linux x86_64/ARM64.
    • New sbx prune command bulk-removes stopped sandboxes with filtering by how long each has been stopped, never touching running sandboxes.
    +9 moreshow less
    • sbx run and sbx create now accept -e/--env and --env-file to inject environment variables into a sandbox, following docker run precedence rules.
    • New --provider ollama flag on sbx run --model selects the Ollama backend, replacing the former ollama/ model-name prefix.
    • sbx secret set and sbx secret set-custom now support dynamic secrets that resolve values from a reference or command, with options to control refreshing, verification, and error output.
    • Sandboxes now expose SANDBOX_NAME and SANDBOX_ID environment variables matching the values shown by sbx ls --json; the older SANDBOX_VM_ID is deprecated.
    • Claude Code's /remote-control endpoint can now be used inside sandboxes by enabling the claude.remoteControl setting via sbx settings set claude.remoteControl true.
    • New platform.images.registryMirror setting redirects Docker Hub-resolving sandbox template and kit images to an organization's registry mirror.
    • Kits can now be signed and verified with cosign-compatible Sigstore signatures via sbx kit sign / sbx kit verify, with optional policy enforcement at load time.
    • OAuth kits can now declare their credential file using the declarative credentialFile.structure form (rendered to well-formed JSON) instead of a free-form Go template.
    • sbx diagnose now reports free disk space on the volume holding sandbox data, detects broken/shadowed/stale SSH client configuration, and checks host virtualization support.
    └──▷ BREAKING ON UPGRADE
    • !sbx run --model no longer accepts an ollama/ prefix to select the Ollama backend; use the new --provider ollama flag instead.
  2. v0.38.0 Aug 6, 2026 · issue 005

    Docker Sandboxes v0.38.0 adds MCP gateway management, kit spec v2, GPU passthrough, per-sandbox network deny rules, and local GGUF model support.

    └──▷ GET THIS VERSION
    $ git clone --branch v0.38.0 https://github.com/docker/sbx-releases.git
    # already have the repo? check out this version:
    $ git checkout v0.38.0
    └──▷ TRY IT
    Block a sandbox from reaching an internal IP range at creation time, getting a clear policy message instead of a silent timeout.
    $ sbx run --deny-network 10.0.0.0/8 claude
    Start a Claude Code session using a local GGUF model served by Ollama instead of the cloud API.
    $ sbx run --model ollama/llama3 claude
    • Adds sbx mcp subcommand for registering and managing remote or local MCP servers, with a built-in MCP gateway that keeps OAuth credentials on the host.
    • Adds dynamic MCP tools mcp-find, mcp-add, and mcp-config-set for attaching registered MCP servers to sandboxes.
    • Adds --deny-network HOST flag to sbx run and sbx create to record per-sandbox network deny rules at creation time, with layer-aware egress messages.
    • Adds sbx daemon restart subcommand to stop and restart the sandboxd daemon in the background.
    • Introduces kit spec schemaVersion: '2' (v2 grammar) with clearer structure for setup, permissions, agent instructions, networking, and credentials; v1 kits continue to load via the legacy path.
    +19 moreshow less
    • Adds DOCKER_SANDBOXES_CLONED_WORKSPACE_SIZE environment variable to configure the size of the cloned workspace volume.
    • Experimental: adds sbx run --gpu for NVIDIA VFIO GPU passthrough on Linux, enabled via sbx settings set feature.sandbox-gpu true.
    • Experimental: adds sbx run --model <name> claude to run Claude Code against a local GGUF model; prefix with ollama/ to use a model from an existing Ollama installation.
    • Experimental: adds proxy, proxy.sandbox, proxy.daemon, and matching no_proxy settings for separate proxy configuration scopes for sandbox and daemon traffic.
    • Experimental: adds proxy.integratedAuth setting on Windows to enable NTLM/Kerberos/Negotiate authentication to upstream proxies.
    • Experimental: adds tls.allowNegativeSerial setting (via sbx settings set tls.allowNegativeSerial true) for compatibility with TLS-inspecting proxies that issue certificates with negative serial numbers.
    • MCP server and tool registrations can be governed organization-wide using Cedar policies.
    • sbx inspect now displays custom secrets configured for a sandbox.
    • sbx create and sbx run show detailed structured startup progress, including environment files loaded, resources provisioned, and each kit command outcome; kit-install progress streams live during sbx create --kit.
    • Blocked HTTPS proxy connections now appear in sbx policy log even when the client aborts the TLS handshake.
    • Docker Agent and OpenCode sandboxes can authenticate GitHub Copilot requests with proxy-managed GitHub credentials.
    • Codex sandboxes created from the TUI prefer stored OpenAI OAuth credentials over API keys.
    • Kit install commands can now consume static files from files/home, including binary files.
    • Kits using extends correctly inherit and override the base image or build source of their parent.
    • Homebrew installs from a stapled .dmg artifact instead of a .tar.gz archive, improving Gatekeeper compatibility on macOS.
    • sbx policy allow network reports a clear error when org governance overrides a local allow rule.
    • Signing in refreshes organization policies in the running daemon immediately instead of waiting for the next polling interval.
    • IP-literal destinations denied by a CIDR rule now fail fast with a policy message instead of timing out.
    • Service and custom secrets are global by default, with --sandbox for sandbox scope; legacy positional and --global forms are deprecated with warnings.
my-toolchain — 0 tools
paste an install list to detect your tools

A brew list, a Brewfile, requirements.txt, a Dockerfile — or just the product names, free-form. Nothing leaves your browser.

    browse all tools →