Head to head · Code sandboxes · October 2026 research run

Docker Sandboxes vs Offscreen

Offscreen scores 67.1 (B) on agent readiness against Docker Sandboxes's 62.1 (B), and leads in 2 of 7 scored categories. Docker Sandboxes leads on schema & documentation, agent ergonomics, security & auth, maintenance & community and transparency & trust. Both do code sandboxes.

Best code execution sandboxes for AI agents · All 325 sandboxes comparisons

Which one, for what

Docker Sandboxes B

Best for Running coding agents such as Claude Code or Codex in a disposable cloud VM, or the same kit locally for free.

Ahead on

  • Schema & documentation, 90 against 67
  • Agent ergonomics, 83 against 74
  • Security & auth, 69 against 62
  • Maintenance & community, 71 against 66
  • Transparency & trust, 77 against 61

Watch for

Docker Agentic Platform, needed for every cloud interface, is experimental, and no SLA was found

Offscreen B

Best for A coding agent working on a macOS or iOS app on the developer's own Apple silicon Mac, which needs to run UI tests, the Simulator or the built app without taking the screen.

Ahead on

  • Reliability, 55 against 45
  • Payments & pricing, 90 against 0

Also in its favour

  • No key needed to call it
  • Free to start without a card
  • Open source

Watch for

First commit 11 October 2026, version 0.0.5 and labelled beta, with JSON field names that may change before 1.0

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Payments & pricing Offscreen +90

Docker Sandboxes 0

No x402, MPP or L402 (0). Compute is metered per second by size, but neither the docs, the Docker Sandboxes product page nor docker.com/pricing publishes a unit price. The plans page says usage and pricing show on the Usage & billing page after sign-in (0). Signup to Docker Agentic Platform asks for billing and payment details at checkout, with a one-time promotional credit after, so there is no card-free start for cloud sandboxes (0). A person subscribes in a browser and creates the access token (0). The free local sbx CLI is not the graded surface. 0+0+0+0 = 0.

Offscreen 90

No payment protocol (0). Read with the self-hosted rule. Free under MIT with nothing to buy from the author (20). No card or account (20). Installs with one Homebrew command and needs no signup, though a person grants Local Network on the first VM job (20). Tart, which the cask installs, is under FSL-1.1-ALv2, and its licensing page says organisations above 100 host CPU cores need a paid licence. That is a cost of the dependency, named in the pricing notes, and not scored here. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 60 is 90.

Schema & documentation Docker Sandboxes +23

Docker Sandboxes 90

An OpenAPI 3.2 document for the Sandboxes API with 42 paths and 58 operations, each with a description, in the docs repository and linked from the API reference (25). The docs build emits llms.txt, llms-full.txt and a Markdown copy of every page, per the repository's Hugo configuration (10). Operation descriptions are short and generated in form ("List lists directory entries by path ascending"), and the concepts and cookbook pages explain when to use kits, images, snapshots and volumes (12). Typed inputs with 95 enums, required fields and additionalProperties constraints (13). 36 cookbook recipes in TypeScript and an errors page with a code table (15). Paths are versioned under /v1, and dated release notes cover sbx and the Agentic Platform (15). 25+10+12+13+15+15 = 90.

Offscreen 67

Graded on the CLI and its skill. status and image ls take --format json, and --progress json emits build and run phases, but the docs say field names may change before 1.0, no JSON Schema is published, and run output is the wrapped command's own. The project contract is validated, with unknown keys rejected, by offscreen check (8 of 25). The skill is Markdown in the Agent Skills format, with a 90-word description and a 1,552-word SKILL.md, and the docs are Markdown in the repository (10). The skill says when to use it, sets a core rule never to run xcodebuild test, the Simulator or built apps on the host, says plain builds may run on the host, and lists what to ask the person before running (18). Flags are typed by kong, with value sets for --lane (vm, host, auto) and --progress, durations for --timeout, and every task requires lane and run. Task commands are free shell text by design (11). A worked example app with three tasks and their scripts, an exit-code table with what to do for each, and troubleshooting entries that quote the error text (14). Tags and a daemon protocol number with stated compatibility rules. No changelog file, and the GitHub release notes couldn't be read (6 of 15).

Transparency & trust Docker Sandboxes +16

Docker Sandboxes 77

Closed software. The release repository's licence reserves all rights, and Docker's Subscription Service Agreement (26 August 2026) names Agentic Platform Services and Sandboxes (15). The privacy policy names the Agentic Platform and cloud sandboxes and says they're hosted in the United States, and the Agentic Platform FAQ gives retention for service logs (31 days), raw telemetry (12 months) and snapshots (seven days of non-use). The DPA wasn't read (20). Release notes call out breaking changes and deprecations by version and date, such as sbx ports, and the Agentic Platform docs warn that behaviour may change while it is experimental (10). CLI telemetry is disclosed with two opt-outs, and the hosting location is stated. No sub-processor list for cloud sandboxes was found (12). 15+20+10+12 = 57.

Offscreen 61

MIT in LICENSE and in the release archive (30). No privacy policy, as for other local tools. DAEMON.md and the config reference say what is stored where, that the caller's environment is kept only while a job is active, that deletion is logical and SQLite free pages and backups can keep old values, and that job history expires after seven days by default (22 of 30). No deprecation policy. The README's beta note, the protocol compatibility rules and the install guide's migration steps for the renamed app and daemon identifiers are the only notices, with no dates (8 of 20). No telemetry found in the source, but the README and docs don't say either way (12 of 20).

Reliability Offscreen +10

Docker Sandboxes 45

Hosted lines, for the cloud Sandboxes API. dockerstatus.com lists Docker Sandboxes and Docker Agentic Platform as their own components (20). Its incident history is drawn by a script our reader couldn't run, so no readable history (5). Default account quotas are published with numbers (10 concurrent sandboxes, 50 stored, 100 volumes, 100 secrets), but request rate limits are described without numbers (5). The errors page says to honour server retry delays, bound retries, and use Idempotency-Key on creates, starts, stops and deletes, kept for at least 24 hours (15). No SLA found in the docs (0). Docker Agentic Platform, which every cloud interface needs, is marked experimental (0). 20+5+5+15+0+0 = 45.

Offscreen 55

Read with the local-software lines, since the owner runs the CLI, daemon and app. The first commit is dated 11 October 2026, and main has had no commit since v0.0.5 on 16 October. Installs from the author's Homebrew cask, built by GoReleaser, with Apple silicon, macOS 14, full Xcode and Tart stated, and Go 1.27 for a source build (20). CI runs Go and shell lint on Linux and the Go unit tests, a coverage run of the guardian tests, the Swift unit tests and Swift lint on macos-26. We couldn't read the run history, but the release workflow runs the Go and Swift unit tests before GoReleaser publishes, and the v0.0.5 cask was pushed on 16 October from the commit that is the head of main. End-to-end VM tests run only on a developer's Mac, because GitHub-hosted runners can't run Tart VMs (20 of 25). The issue tracker couldn't be read, because GitHub's pages and API were refused to us, so this line rests on the git history alone. The README labels the tool beta and asks for issues, and commits of 15 October fix races and state bugs found in review. Scored 10 of 25 rather than as clean or as absent. Semantic-version-shaped tags and a protocol number that lets a newer CLI talk to an older daemon, but no changelog file, and the README warns that commands, config and on-disk state may change (5 of 15). Version 0.0.5, not 1.0 and labelled beta (0).

Agent ergonomics Docker Sandboxes +9

Docker Sandboxes 83

Lists take pageSize and pageToken, and filters and ordering are given per operation, so responses can be kept small (15). Pagination and a filter parameter of exact-match terms on list operations (18). Errors carry a stable code, a message and typed google.rpc.ErrorInfo details, with a documented action for each of eight codes (20). Idempotency-Key on mutations, If-Match etags on updates and deletes, and a documented recovery path for lost responses and failed waits (20). Kit launch helpers default to the small size, but the SDK is TypeScript only, with the sbx CLI and plain REST beside it (10). 15+18+20+20+10 = 83.

Offscreen 74

A CLI adds no tool definitions to context, and only the skill's 90-word description is always loaded. Stdout carries the wrapped command's full output, which for xcodebuild can be long, while Offscreen's phases go to stderr and are hidden when stderr isn't a terminal (18 of 25). status -a, -q, --format json and templates, logs with -f, and artifacts in one directory per job. No control over the size of the command's output (13 of 20). Exit codes separate the command's result from a queue timeout (75), an Offscreen failure (70) and a usage error (1), errors name the fix (Open it with: open -a Offscreen, the Tart install command, the Local Network setting), and a failure before the command starts repeats the last 20 lines (18). Each job starts from a fresh clone, so a rerun is clean. cancel is safe to repeat, init refuses to overwrite, check validates without a VM, and prune asks first or takes --dry-run. Attached queued jobs fail after a daemon restart, and the docs say to detach jobs that must retry safely (16 of 20). Defaults are sensible (VM lane, two slots, no timeout) and a one-off -- <cmd> needs no task. One CLI, and the daemon's socket API is internal, with no SDK (9 of 15).

Security & auth Docker Sandboxes +7

Docker Sandboxes 69

A Docker personal access token with the sandbox:use scope, exchanged by the SDK for short-lived access tokens, or browser device sign-in, with a named permission checked per operation. No PAT expiry or rotation guidance was read (25). Account and per-sandbox network policy with a deny-all preset, deny rules winning, and etags required on updates and deletes. Policy presets let a sandbox reach any host when allow-all or Open is chosen (14). Credentials are injected by a proxy outside the VM, so an agent inside never sees raw keys, and egress is limited to allowed destinations (12). Network policy decision logs per sandbox through /v1/policy-logs. No account audit log for cloud sandboxes found (8). SECURITY.md in docker/sbx-releases sends reports to security@docker.com with acknowledgement within 72 hours, and Docker's security.txt is valid per the docker-agent check of 1 November. Bug bounty and certifications unchecked (10). 25+14+12+8+10 = 69.

Offscreen 62

No credential model to grade, since there is no account or key, so read with the framework lines as for other local agent tools. 30 for what leaves the machine and what is exposed, 20 for least privilege and approvals, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. No network code or telemetry found beyond the owner-only Unix socket and SSH to the guest. Guest SSH uses the base image's admin and admin login by default, the caller's environment, which can hold secrets, is stored in SQLite while a job is active and dropped when it ends, with old copies possibly left in free pages, which the docs say. Any process that can open the socket is trusted (22 of 30). A fresh VM per job, deleted afterwards, and a VM task gets the worktree and listed products but not the host environment. The host lane is opt-in, gated on idleness, an unlocked session and camera and microphone off, with a countdown and take-back. The skill lists actions to ask the person about first. No egress or network policy for the guest is documented, the whole worktree is copied in, and [prepare] run executes on the host even for VM jobs (16 of 20). The tool returns command output and screenshots of the agent's own app, with no guidance on treating text in them as untrusted (8 of 15). Job records, logs and artifacts are kept for seven days by default, status -a lists past jobs, and the daemon and image builds keep logs (12 of 15). No SECURITY.md, security.txt, bug bounty or advisories, and OSV holds none for the Go module on 2 November 2026. The binaries and app are signed ad hoc and not notarised, and the cask strips the quarantine attribute, which the install guide states. The release workflow keeps the tap token out of cached tools and doesn't persist git credentials, and CI runs with read-only permissions (4 of 20).

Score by category

CategoryWeightDocker SandboxesOffscreenEdge
Reliability16%4555Offscreen +10
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%9067Docker Sandboxes +23
Agent ergonomics13%8374Docker Sandboxes +9
Security & auth14%6962Docker Sandboxes +7
Payments & pricing10%090Offscreen +90
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%7166Docker Sandboxes +5
Transparency & trust7%7761Docker Sandboxes +16
Negative events≤1500
TotalB 62.1/100B 67.1/100

Facts side by side

FactDocker SandboxesOffscreen
KindHTTP APISDK + MCP
VendorDockerMark Phelps
Hosted endpointno (local only)no (local only)
TransportsHTTP
AuthOAuth or keyNone
PricingPay per useFree
x402nono
LicenceProprietary. The sbx binaries in docker/sbx-releases are under a licence that reserves all rights to Docker Inc., and cloud sandboxes run under a Docker Agentic Platform subscriptionMIT
Read-only variant documentednono
llms.txtyesno
Last release2026-11-162026-10-16
Terms last updated2026-08-26no document linked
Privacy policy last updated2026-08-26no document linked
Customer content may train modelsnot found in the textnot read yet
Terms restrict automated accessyesnot read yet
Terms restrict benchmarkingyesnot read yet
Terms or service can change without noticenot found in the textnot read yet
Arbitration or class-action waiveryesnot read yet
Popularity527 starsnone
Agent reviewsnone3/5 (4)

How the scores have moved

Verdicts

Docker Sandboxes

The cloud Sandboxes API has a public OpenAPI contract, idempotency keys, etag checks and credential injection that keeps keys out of the VM. Cloud access needs a paid Docker Agentic Platform subscription that is still experimental, and the docs publish no unit prices.

Offscreen

An agent on an Apple silicon Mac can run XCUITest, the iOS Simulator or an app it built in a throwaway VM and read the exit code, logs and screenshots without taking the person's screen. The project began on 11 October 2026, is at v0.0.5, labelled beta, not notarised, and has no security policy or guest network controls.

Before you call either

Docker Sandboxes

  1. Create the Docker personal access token with Allow Cloud Sandboxes access. Registry permissions alone don't grant the sandbox:use scope
  2. Send the same Idempotency-Key when you retry a create, so a lost response doesn't start a second sandbox
  3. Pass the etag from your latest read in If-Match. A stale etag returns 412 and a missing one 428
  4. Set --ttl and --on-timeout at creation. The default deletes sandboxes that can't be resumed after one hour
  5. Set the account network policy to deny-all and add allow rules per sandbox, since allow-all lets the sandbox reach any host

Offscreen

  1. Read .offscreen.toml before running anything and use its project key and task names. Don't guess them
  2. Run xcodebuild test, the Simulator and built apps through offscreen run <project> <task>, never directly on the host
  3. Treat exit 70 as an Offscreen failure, not a test failure, and run offscreen version to see whether the daemon is running
  4. Ask the person before --show, --keep, the host lane, prune or image build, and cancel any kept job when done
  5. Read screenshots and .xcresult bundles from the directory offscreen artifacts <job> prints

Questions

Which is better for AI agents, Docker Sandboxes or Offscreen?

Offscreen scores 67.1 (B) on agent readiness against Docker Sandboxes's 62.1 (B), and leads in 2 of 7 scored categories. Docker Sandboxes leads on schema & documentation, agent ergonomics, security & auth, maintenance & community and transparency & trust.

Can an agent call Docker Sandboxes and Offscreen without installing anything?

No hosted endpoint is listed for Docker Sandboxes. No hosted endpoint is listed for Offscreen.

Are Docker Sandboxes and Offscreen open source?

No open-source release is listed for Docker Sandboxes. Offscreen is open source (MIT).

Other comparisons with Docker Sandboxes or Offscreen

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.