Category · Agent runtime
Code execution sandboxes for AI agents
Isolated machines an agent can start in seconds to run code, install packages and use a filesystem, then throw away. Compared on start time, isolation, how long a sandbox can live, what it costs per second and whether state can be paused and resumed.
Best code execution sandboxes for AI agents, the shortlist231 head-to-head comparisons
Capability keys sandbox.code · sandbox.fs · sandbox.persist · sandbox.browser · sandbox.gpu · All tools
letme.dev/sandbox.code picks the top-graded tool in this list and says how to call it direct (how letme works).
The same listing from the live API. Graded results come first, then the official MCP registry when no graded-only filter is set.
https://www.anchorterminal.com/api/v1/search
Filters
| Compare | # | Tool | Category | Grade | Score | Agent rating | p95 | Context | Price / x402 | Auth | Where | Details |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 77 | Microsoft Execution ContainersMicrosoft · SDK + MCP | Sandboxes | BB | 77.1 | 3.0 (4) | n/a | n/a | Free · OSS | None | Local | ||
|
Microsoft Execution Containers (MXC) is an open-source SDK for running untrusted code in a local sandbox on Windows, Linux and macOS. An application embeds it through Node.js, .NET or Rust and sets filesystem, network and UI policy for each run. Top strength MIT licence, with SDKs for Node.js, .NET and Rust all at 1.0.1 and the native runtime bundled in the npm and NuGet packages Top weakness 1.0.1, the newest release on 1 November 2026, still grants read access to the whole system drive from the Windows tool-discovery helper when PowerShell is on PATH, fixed on main only |
||||||||||||
| 90 | Vercel SandboxVercel · HTTP API | Sandboxes | BB | 76.6 | 3.5 (2) | n/a | n/a | $0.128 / vCPU-hr | OAuth or key | Hosted | ||
|
Firecracker microVM sandboxes on Vercel, driven from the `@vercel/sandbox` JavaScript SDK, the Python `vercel` package, a CLI or the REST API. Top strength Active CPU billing, so waiting on model responses costs only memory Top weakness Tied to a Vercel team and project even when called from elsewhere, and access tokens reach the whole team |
||||||||||||
| 187 | E2BE2B · HTTP API | Sandboxes | BB | 73.2 | 3.0 (2) | n/a | n/a | $0.0504 / vCPU-hr | API key | Hosted | ||
|
Firecracker microVM sandboxes for agent code, driven from Python and JavaScript SDKs, a CLI or a REST API. Top strength Firecracker microVM with its own kernel per sandbox Top weakness Minor SDK releases 2.53.0, 2.56.0 and 2.62.0 (October 2026) shipped changes marked breaking, with no earlier notice |
||||||||||||
| 218 | Amazon Bedrock AgentCore Code InterpreterAmazon Web Services · HTTP API | Sandboxes | BB | 72.4 | 3.3 (4) | n/a | n/a | $0.0895 / vCPU-hr | API key | Hosted | ||
|
Amazon Bedrock AgentCore Code Interpreter is AWS's managed sandbox for running agent-written Python, JavaScript and TypeScript. Each session is a dedicated microVM, reached through the AWS API, the AgentCore SDKs or an MCP server. Top strength Each session runs in a dedicated microVM, which AWS says is terminated and has its memory sanitised when the session ends Top weakness No pause, resume or snapshot. Session files are removed when the session ends, and persistence needs a customer-owned S3 Files or EFS mount inside a VPC |
||||||||||||
| 252 | Celesto (formerly SmolVM)Celesto AI, Inc. · SDK + MCP | Sandboxes | BB | 71.8 | none | n/a | n/a | $0.015 / session-hr | None | Local | ||
|
Celesto, formerly SmolVM, is an open-source Python SDK and CLI from Celesto AI that runs disposable Firecracker or QEMU microVMs on the owner's machine for agent code, browsers and desktops, with an optional paid Celesto Cloud. Top strength Apache-2.0, installs from PyPI with Top weakness Version 0.1.13 and declared pre-1.0 in SECURITY.md, with the TypeScript SDK an alpha installed from a GitHub tarball |
||||||||||||
| 259 | Cloudflare Sandbox SDKCloudflare · SDK + MCP | Sandboxes | BB | 71.6 | 3.0 (2) | n/a | n/a | $0.072 / vCPU-hr | None | Local | ||
|
TypeScript library for running sandboxed Linux containers from a Cloudflare Worker. Top strength Each sandbox runs in its own VM with a separate filesystem, process space and network stack Top weakness No hosted API. You deploy and secure a Worker before an agent can call anything, and the starter has no auth |
||||||||||||
| 284 | DaytonaDaytona · HTTP API | Sandboxes | BB | 70.8 | 3.0 (2) | n/a | n/a | $0.0504 / vCPU-hr | API key | Hosted and local | ||
|
Sandboxes for agent code in container, Linux VM, Windows and GPU classes, driven by SDKs for Python, TypeScript, Ruby, Go and Java or a REST API. Top strength API keys with per-action scopes, so an agent can create sandboxes without being able to delete them Top weakness The container class shares the host kernel. Only the VM classes get their own |
||||||||||||
| 354 | MaritimeMaritime AI, Inc. · HTTP API | Sandboxes | B | 68.9 | 4.0 (4) | n/a | n/a | $20 / mo | API key | Hosted | ||
|
Maritime hosts long-lived AI agents, each in its own micro-VM with a persistent disk that sleeps when idle and wakes in about a second. Agents and cloud computers are driven through a REST API, a CLI, two SDKs or MCP. Top strength Public OpenAPI file, llms.txt and every docs page in Markdown. Top weakness No status page or SLA. |
||||||||||||
| 401 | SpritesFly.io · HTTP API | Sandboxes | B | 68 | none | n/a | n/a | $0.0385 / vCPU-hr | API key | Local | ||
|
Sprites are hosted Linux microVMs from Fly.io that keep their disk between runs, pause when idle and wake on request. They are driven through a REST API, a CLI, four SDKs or a hosted MCP server. Top strength The filesystem persists across pauses, and compute billing stops about 30 seconds after activity ends, per the lifecycle docs Top weakness Five partial Sprites incidents on the status page since 30 August 2026, including 7 hours 28 minutes of intermittent API failures on 23 September and 75 minutes of unreachable Sprite URLs on 11 October |
||||||||||||
| 436 | Modal SandboxesModal · SDK + MCP | Sandboxes | B | 67.3 | 3.3 (8) | n/a | n/a | $0.071 / vCPU-hr | API key | Local | ||
|
Modal's sandboxed compute environments for running code, with SDK access, GPU support and filesystem snapshots. Top strength GPU sandboxes at the same per-second rates as the rest of Modal Top weakness No REST API, and the JavaScript and Go SDKs are beta |
||||||||||||
| 446 | OffscreenMark Phelps · SDK + MCP | Sandboxes | B | 67.1 | 3.0 (4) | n/a | n/a | Free · OSS | None | Local | ||
|
Offscreen is an open-source macOS command-line tool, daemon and menu bar app that runs a coding agent's UI tests, iOS Simulator runs and GUI app launches in a fresh Tart virtual machine per job, away from the user's desktop. Top strength A fresh VM per job, deleted afterwards, and a VM task receives only the worktree, listed build products and four Top weakness First commit 11 October 2026, version 0.0.5 and labelled beta, with JSON field names that may change before 1.0 |
||||||||||||
| 630 | Runloop DevboxesRunloop · HTTP API | Sandboxes | B | 63.2 | 3.0 (2) | n/a | n/a | $0.108 / vCPU-hr | API key | Hosted | ||
|
Devboxes, VM sandboxes for coding agents, with blueprints, disk snapshots, suspend and resume, idle policies and a credential gateway. Runloop is winding down its services, with support for existing customers through 13 November 2026. Top strength Agent gateways keep real credentials on Runloop's servers, with gateway tokens bound to one devbox Top weakness Runloop is winding down, and support for existing customers ends on 13 November 2026, while runloop.ai still invites free sign-ups with no notice |
||||||||||||
| 672 | FreestyleFreestyle Cloud Inc. · HTTP API | Sandboxes | B | 62.4 | none | n/a | n/a | $0.0403 / vCPU-hr | API key | Hosted | ||
|
Freestyle runs full Linux virtual machines for AI agents, with pause and resume that keeps memory, snapshots, private networks and domains. Agents drive it through a REST API, a TypeScript SDK and a CLI from one npm package. Top strength Public OpenAPI 3.1 file with 88 Top weakness No terms of service or service agreement found on the site, and the privacy policy covers the website only |
||||||||||||
| 695 | Blaxel SandboxesBlaxel · HTTP API | Sandboxes | C | 61.8 | 2.0 (2) | n/a | n/a | $0.1656 / session-hr | OAuth or key | Hosted | ||
|
Sandbox VMs that drop to standby seconds after the last connection and resume in about 25 ms with memory and filesystem kept, charging only for snapshot storage while idle. Top strength No compute charge in standby, only $0.20 a GB-month of snapshot storage Top weakness 32 status-page incidents from 6 August to 3 November 2026, 31 touching Sandboxes, each limited to a region, a capability or some users |
||||||||||||
| 755 | DeltaBox LiteDeltaBox (delta-box) · SDK + MCP | Sandboxes | C | 60.4 | 2.8 (4) | n/a | n/a | Free · OSS | None | Local | ||
|
DeltaBox Lite is an open-source local sandbox runtime for AI coding agents on Linux, macOS and Windows. Each command runs in a confined view of the project, and its file changes can be diffed, committed or aborted. Top strength Apache-2.0, free and local, installing from npm, Homebrew or checksummed release tarballs for Linux and macOS on x64 and arm64 and for Windows x64 Top weakness Version 0.1.5, first released on 14 October 2026, with six releases and no stability declaration |
||||||||||||
| 812 | SPACEPerplexity · HTTP API | Sandboxes | C | 58.9 | 3.0 (4) | n/a | n/a | Free · OSS | OAuth or key | Local | ||
|
SPACE is Perplexity's open-source, self-hosted platform for running agent code in Firecracker microVMs. An operator deploys it on Linux with KVM, and applications drive sandboxes through its HTTP and SSE API or an E2B-compatible API. Top strength OpenAPI 3.1 for 14 supported operations, served by each deployment at Top weakness A person must deploy it first, on an apt-based Linux x86_64 host with KVM, privileged Docker and Kubernetes (Kind in the guide) |
||||||||||||
| 907 | Sandbox0Lougao, LLC (Sandbox0) · HTTP API | Sandboxes | C | 56.6 | 3.0 (4) | n/a | n/a | $0.015 / unit | API key | Hosted | ||
|
Sandbox0 Cloud runs managed gVisor sandboxes for AI agents whose whole writable root filesystem persists across pause and resume, with snapshots, forks and egress controls, reached through a REST API, Go, Python and TypeScript SDKs and the `s0` CLI. Top strength Whole writable root filesystem persists across pause, with snapshots, forks and file access while paused Top weakness No status page, incident history or SLA found |
||||||||||||
| 946 | Together Code SandboxTogether AI · HTTP API | Sandboxes | C | 56 | none | n/a | n/a | $0.0446 / vCPU-hr | API key | Hosted | ||
|
Together AI's hosted virtual machine sandboxes for running commands and code, built from Docker-image snapshots and driven from the `together-sandbox` Python SDK, TypeScript SDK or CLI. Access is by allowlist, on request to Together. Top strength Public OpenAPI documents for the management API (17 operations) and the in-VM API (25), in the repository and on GitHub Pages Top weakness The SDK and CLI work only for organisations Together has enabled, and the docs say to contact Together for access |
||||||||||||
| 953 | Agent 37 CloudAgent 37 Inc. · HTTP API | Sandboxes | C | 55.7 | 2.8 (4) | n/a | n/a | $0.0011 / vCPU-hr | API key | Hosted | ||
|
Agent 37 Cloud hosts persistent Linux sandboxes for agent harnesses such as Hermes, OpenClaw, Claude Code and Codex, or a custom Docker image. A REST API creates instances, runs commands, manages files and sends messages to the agent inside. Top strength Disk persists until delete, and auto-sleep checkpoints an idle instance so that only disk bills ($0.09 a GB-month) Top weakness The status page is new, with history only from October 2026, and the terms promise no uptime level |
||||||||||||
| 1028 | InstaVMChecksum Labs, Inc. · HTTP API | Sandboxes | D | 53.4 | none | n/a | n/a | $0.0504 / vCPU-hr | API key | Hosted | ||
|
InstaVM is a hosted service from Checksum Labs that runs AI agents' code, browsers and desktops in Firecracker microVMs, with snapshots, volumes, egress policies and SSH, through a REST API and Python and JavaScript SDKs. Top strength Public OpenAPI 3.1 at api.instavm.io/openapi.json with 229 operations, plus llms.txt Top weakness The free plan's $50 credit needs a card on file, and signup is in a browser |
||||||||||||
| 1052 | Morph CloudMorph Labs · HTTP API | Sandboxes | D | 52.8 | none | n/a | n/a | $0.05 / unit | API key | Hosted | ||
|
Cloud virtual machines for AI agents from Morph Labs. Instances boot from snapshots and can be paused with memory intact, branched into copies, and driven through a REST API, Python and TypeScript SDKs and a CLI. Top strength Pause and snapshot preserve memory and running processes, and a paused instance is billed for its snapshot only Top weakness No privacy policy, DPA or subprocessor list found on morph.so or cloud.morph.so. The terms of 23 October 2025 are the only legal document |
||||||||||||
| 1121 | Deno SandboxDeno Land Inc. · HTTP API | Sandboxes | D | 50.8 | none | n/a | n/a | $0.10 / vCPU-hr | API key | Local | ||
|
Deno Sandbox runs Linux microVMs on Deno Deploy for untrusted or AI-generated code, through JavaScript and Python SDKs or the `deno sandbox` CLI. In beta since February 2026, it closes with Deno Deploy six months after 9 October 2026. Top strength Secrets are held outside the VM. Code sees a placeholder, and the real value is substituted only on requests to hosts named for that secret Top weakness Deno Deploy, which runs Sandbox, shuts down six months after the 9 October 2026 announcement, and Deno is moving its development to Cloudflare |
||||||||||||
Nothing matches these filters. .
p95 latency and context cost come from our probes, which haven't run yet, so those columns start hidden. Grades run from AA to F, and agent-ready means BB or better. Filters, sorting and export run in your browser; the table is complete without JavaScript.
Indexed, not reviewed (11)
Listings sorted into this category from public catalogues (the official MCP registry, APIs.guru, the x402 Bazaar and OpenRouter), with facts and our own checks but no score, grade or rank. How the index works.
| Listing | Kind | What it does | Why it's here |
|---|---|---|---|
| Auth Your Agent authyouragent.com | MCP server | Act for a person on websites: sandboxed browser, phone approvals, saved logins, scheduled plans. | vendor's own |
| Errata formachines.ai | MCP server | Sandbox-verified fixes where AI models write outdated code for PyPI and npm libraries | vendor's own |
| Kenwea — Sandbox Attestation & Agent Marketplace www.kenwea.com | MCP server | Signed sandbox verdicts on any artifact, plus an agent marketplace. No key, no signup, no payment. | vendor's own |
| ParallelSandbox parallelsandbox.com | MCP server | A row of in-progress test builds of your app. Use each one, say what's wrong, circle it on screen. | vendor's own |
| Pink Agentic AI Payments pinkwallet.com | MCP server | Pink Agentic AI Payments: AI agents pay within spending rules (budgets, caps, approval). Sandbox. | vendor's own |
| Qut Pay (Kaspi QR payments) qut.kz | MCP server | Kaspi QR payments (Kazakhstan) via Qut Pay: create invoices, check status, refund, sandbox tests. | vendor's own |
| Rivet rivet.dev | MCP server | Manage Rivet Cloud namespaces and actors, run Code Mode, and open the Rivet Actor Inspector. | vendor's own |
| Runtime Cloud withruntime.com | MCP server | Linux microVM sandboxes for AI agents: run commands, files, processes, pause and wake. | vendor's own, widely used |
| SandboxAPIs sandboxapis.dev | MCP server | Drop-in read-only replicas of GitHub, Jira, Slack and 22 more, preloaded with one fake company. | vendor's own |
| Sandboxer sandboxer.com | MCP server | Cloud Linux VMs for your agent: run code, Docker, files and HTTPS previews in a Sandboxer microVM. | vendor's own |
| ScratchRun scratchrun.dev | MCP server | Ephemeral MicroVM-isolated code execution for AI agents. Fresh VM per call, hard-purged after. | vendor's own |
How we test this category
The same task in every sandbox: start, install a package, run a script that writes files, pause and resume where supported, then tear down. We time each step, check isolation claims against the docs and add up the cost.
How the ranking works
Every tool is scored 0 to 100 across 9 weighted categories, minus deductions for negative events, and given a grade from AA to F. Reliability, latency and context cost are measured by probes and by traffic through letme. Schema quality, security and payments are assessed against a published checklist. Task success comes from a fixed task suite. Payments, including x402, are one category of the nine. Read the methodology.
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.