Alternatives · Agent sandboxes · verified 17 Oct 2026

Amazon Bedrock AgentCore alternatives

Teams look past AgentCore when the agent they want is Claude Code or Codex, and the work should end in a pull request rather than an API response they assemble into one.

Facts checked 17 October 2026. Prices and limits move. Read the citations.

Start a TaskAll comparison pages
RepositoryGitHub remoteSandboxOne per TaskHarnessClaude · Codex · OpenCodeTrajectoryKeptPRor review
Isolated sandbox rooms, one per run

Rush is our product and is listed first. Every other row is from the vendor's own pages, retrieved on the date beside it.

Read this first

The useful distinction

Teams look past AgentCore when the agent they want is Claude Code or Codex, and the work should end in a pull request rather than an API response they assemble into one.

A sandbox is a computer for an agent. You still write the agent loop, clone the repository, push the branch and decide when the work is done. Rush is the layer above: you send the task and get back the pull request, report or files.

Routing

If you… use… not if

If youUseNot if
You want to send a task by API or MCP and get back a pull request, report or files with the transcript.RushYou need to self-host, use GitLab or Bitbucket, or bring an agent loop of your own.
You are building your own agent product and want raw microVM sandboxes with an SDK.E2BYou want the task done and a pull request back without writing the loop.
You need many sandbox shapes (Linux, Windows, macOS, GPU) under one SDK, or your own runners.DaytonaYou want an agent to finish the task and hand back a pull request.
You need GPUs in the sandbox or very large fan-out for RL rollouts and evals.Modal SandboxesThe job is one agent on one repository, ending in a pull request.
You are building an agent platform and need credentials kept out of the sandbox, a VPC deploy or RL fine-tuning.RunloopYou want to send a task and get a pull request without writing orchestration.
You are on Vercel and want sandboxes billed only for active CPU, in 19 regions.Vercel SandboxYou want the task done and the pull request back without a loop of your own.
You are already on Workers, R2 and Durable Objects and want sandboxes next to them.Cloudflare SandboxYou want a pull request back, not a diff to commit yourself.

Landscape

The field

Prices and limits from vendor pages retrieved 17 October 2026. Check the source before you buy.

ToolJobWhat runsIsolationProofPriceTaps out
RushSend a task by API; an agent does it in its own sandboxClaude Code, Codex or OpenCodeOne E2B microVM per runPR with checks, artifacts, transcript; else needs_reviewFree 5 agent-hours / month, then $20 ProThree harnesses. GitHub repos only. No self-host.
Amazon Bedrock AgentCoreAWS platform to host and govern agentsAny framework; Strands-based harnessMicroVM per sessionNone; API response$0.0895 / vCPU-h; $200 credit for new AWS8 h sessions. You assemble the parts.
E2BSDK for microVM sandboxes your agent runs inYour agent (Claude Code template ships)Firecracker microVMNone; you read git diffHobby $0 + $100 credit; Pro $150 / mo + usage1 h Hobby, 24 h Pro sessions. No PR layer.
DaytonaElastic sandboxes for AI-generated codeYour agentContainer by default; VM opt-inNone; you push and open the PR$200 free, then per second15-min auto-stop. 4 vCPU / 8 GB ceiling.
Modal SandboxesServerless sandboxes for agents, RL and untrusted codeYour agentgVisor, or VM runtimeNone; you build itStarter $0 + $30 credit; Team $250 / mo5-min default, 24 h cap per sandbox.
RunloopDevboxes and SDK for software engineering agentsYour agent; Claude Code and ACP adaptersVM per DevboxNone; you open the PRBasic free + $50 credit; Pro $250 / moYou write the loop. VPC on Enterprise.
Vercel SandboxFirecracker sandboxes from Vercel's SDK and CLIYour agent (coding agents in the image)Firecracker microVMNone; you open the PRHobby free 5 CPU-h; Pro from $20 credit24 h sessions on Pro, 45 min on Hobby.
Cloudflare SandboxMicroVM containers started from a WorkerYour agentFirecracker microVMNone; tutorial returns a diffWorkers Paid $5 / mo + per second4 vCPU / 12 GiB max. Worker per sandbox.

Depth

In depth

Hobby $0 with a one-time $100 credit; Pro $150 / month plus usage. Checked 17 October 2026.

E2B

You are building your own agent product and want raw microVM sandboxes with an SDK.

Firecracker microVM sandboxes driven from an SDK. Your agent runs code there; the loop, the git steps and the pull request are your code. Rush itself runs on E2B.

Left to you. E2B's Claude Code guide clones with your token, runs the agent and reads git diff; pushing and opening the pull request is your code.

Taps out. Sessions end at 1 hour on Hobby and 24 hours on Pro unless paused. Sandboxes reach the open internet by default. No result or pull request layer. Pro has a $150 monthly fee before usage.

$200 of free compute, then pay as you go with no monthly fee. Checked 17 October 2026.

Daytona

You need many sandbox shapes (Linux, Windows, macOS, GPU) under one SDK, or your own runners.

Elastic sandboxes for AI-generated code: Linux containers by default, with VM, Windows, macOS and GPU options, SDKs in five languages, an MCP server and BYOC.

Left to you. The git API clones, branches, commits and pushes; opening the pull request is your code.

Taps out. Default runtime is a container, and container sandboxes cannot pause memory. The 15-minute auto-stop does not count a running process as activity. The open-source repository stopped receiving updates in June 2026.

Starter $0 + compute with $30 / month credit; Team $250 / month + compute. Checked 17 October 2026.

Modal Sandboxes

You need GPUs in the sandbox or very large fan-out for RL rollouts and evals.

Programmable gVisor containers, or a VM runtime, on Modal's serverless cloud, for agent code, RL rollouts and untrusted code, with snapshots, egress rules and GPUs.

Left to you. The sandbox guide has no git or pull request feature.

Taps out. 5-minute default lifetime and a 24-hour cap per sandbox. Sandbox CPU costs about three times a Modal function's. GPU sandboxes need the gVisor runtime.

Basic free with $50 of credits; Pro $250 / month plus usage. Checked 17 October 2026.

Runloop

You are building an agent platform and need credentials kept out of the sandbox, a VPC deploy or RL fine-tuning.

VM-isolated Devboxes for software engineering agents, with blueprints, snapshots, an MCP Hub, Agent Gateways that hide API keys, and a Remote Agents SDK.

Left to you. Code Mounts pull a repository or branch in; opening the pull request is your code.

Taps out. You still write the agent loop. Suspend and resume and repo connections are listed under the $250 Pro plan. VPC deploy is Enterprise only.

Hobby free with 5 Active CPU hours / month; Pro draws from the $20 / month credit. Checked 17 October 2026.

Vercel Sandbox

You are on Vercel and want sandboxes billed only for active CPU, in 19 regions.

An SDK, CLI and API for on-demand Firecracker microVMs that run untrusted or agent-generated code. You bring the agent and the workflow. Generally available since 30 January 2026.

Left to you. You get exec, files and logs; the pull request is your code.

Taps out. 5-minute default timeout. 24-hour session cap on Pro, 45 minutes on Hobby. A Pro sandbox tops out at 8 vCPU and 16 GB. Hobby creation pauses once quota is used.

Workers Paid $5 / month, with 375 vCPU-minutes and 25 GiB-hours included. Checked 17 October 2026.

Cloudflare Sandbox

You are already on Workers, R2 and Durable Objects and want sandboxes next to them.

Linux microVM containers, or isolated Dynamic Workers, started from a Worker to run untrusted or agent-generated code. You build the agent loop.

Left to you. Cloudflare's coding agent runner tutorial clones a repository, runs Claude Code and returns a diff, with the instruction not to commit.

Taps out. Needs Workers Paid and a Worker and Durable Object around each sandbox. Largest instance is 4 vCPU and 12 GiB. The tutorial ends at a diff, with no pull request or proof.

Rush

What Rush is

Rush is the alternative when you went to AgentCore to run a coding agent and found you would be assembling the runtime, gateway, identity and the pull request step yourself.

Honest limits

Skip Rush if

  • Rush runs three agents: Claude Code, Codex and OpenCode. It does not host an agent you wrote yourself.
  • Repositories come through the Rush GitHub App. A GitLab or Bitbucket repository cannot be a run's repository yet.
  • There is no self-hosted or customer-VPC deployment, and we do not publish a compliance report today.
  • Compute is metered. Free includes 5 agent-hours a month and Pro 40, each with a weekly ceiling.
  • You review the pull request. Rush does not merge for you.

FAQ

Questions people actually ask

What is the best Amazon Bedrock AgentCore alternative?

It depends on the part of the job you want to keep. Use E2B if you are building your own agent product and want raw microVM sandboxes with an SDK. Use Daytona if you need many sandbox shapes (Linux, Windows, macOS, GPU) under one SDK, or your own runners. Use Modal Sandboxes if you need GPUs in the sandbox or very large fan-out for RL rollouts and evals. Use Rush if you want to send a task by API or MCP and get back a pull request, report or files with the transcript.

Can I move from Amazon Bedrock AgentCore to Rush?

AgentCore is a platform for agents you build, so there is no single thing to move. A custom coding agent on AgentCore Runtime can be replaced by a Rush run; the rest stays on AWS.

Where does the code go when I leave?

Anything pushed to your GitHub remote stays there. That is the off-ramp on every product in this set, including Rush.

Cluster

Keep reading

Citations

Sources

Every factual claim on this page traces to one of these URLs. Retrieval date is 17 October 2026 unless the source prints an earlier published date.

  1. [1] Rush, “Rush product requirements”, retrieved 2026-09-15. https://getrush.ai/cloud
  2. [2] Rush, “Rush pricing (live contract on getrush.ai)”, retrieved 2026-09-15. https://getrush.ai/pricing
  3. [3] Rush, “Isolation, plainly”, retrieved 2026-10-17. https://getrush.ai/docs/concepts/isolation
  4. [4] Rush, “Results and proof”, retrieved 2026-10-17. https://getrush.ai/docs/concepts/results-and-proof
  5. [5] Rush, “Usage and billing”, retrieved 2026-10-17. https://getrush.ai/docs/concepts/usage-and-billing
  6. [6] Rush, “Rush MCP server”, retrieved 2026-10-17. https://getrush.ai/docs/concepts/mcp-server
  7. [7] Rush, “Automations”, retrieved 2026-10-17. https://getrush.ai/docs/concepts/automations
  8. [8] AWS, “Amazon Bedrock AgentCore - AWS”, retrieved 2026-10-17. https://aws.amazon.com/bedrock/agentcore/
  9. [9] AWS, “Amazon Bedrock AgentCore Pricing - AWS”, retrieved 2026-10-17. https://aws.amazon.com/bedrock/agentcore/pricing/
  10. [10] AWS, “Amazon Bedrock AgentCore is now generally available”, published 2025-10-13. https://aws.amazon.com/about-aws/whats-new/2025/10/amazon-bedrock-agentcore-available/
  11. [11] AWS, “Overview (What is Amazon Bedrock AgentCore?)”, retrieved 2026-10-17. https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/what-is-bedrock-agentcore.html
  12. [12] AWS, “Host agent or tools with Amazon Bedrock AgentCore Runtime”, retrieved 2026-10-17. https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/agents-tools-runtime.html
  13. [13] AWS, “AgentCore harness”, retrieved 2026-10-17. https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness.html
  14. [14] AWS, “Interact with web applications using Amazon Bedrock AgentCore Browser”, retrieved 2026-10-17. https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/browser-tool.html
  15. [15] E2B, “E2B Pricing”, retrieved 2026-10-17. https://e2b.dev/pricing
  16. [16] E2B, “E2B | The Enterprise AI Agent Cloud”, retrieved 2026-10-17. https://e2b.dev/
  17. [17] E2B, “Sandbox persistence”, retrieved 2026-10-17. https://docs.e2b.dev/sandbox/persistence
  18. [18] E2B, “Claude Code - E2B Docs”, retrieved 2026-10-17. https://docs.e2b.dev/agents/claude-code
  19. [19] E2B, “Coding Agents - E2B Docs”, retrieved 2026-10-17. https://docs.e2b.dev/use-cases/coding-agents
  20. [20] E2B, “Available servers - E2B Docs”, retrieved 2026-10-17. https://docs.e2b.dev/mcp-gateway/available-servers
  21. [21] E2B, “e2b-dev/infra (GitHub README)”, retrieved 2026-10-17. https://github.com/e2b-dev/infra
  22. [22] Daytona, “Daytona - Secure Infrastructure for Running AI-Generated Code (pricing)”, retrieved 2026-10-17. https://www.daytona.io/pricing
  23. [23] Daytona, “Sandboxes | Daytona”, retrieved 2026-10-17. https://www.daytona.io/docs/en/sandboxes
  24. [24] Daytona, “Daytona MCP Server | Daytona”, retrieved 2026-10-17. https://www.daytona.io/docs/en/mcp
  25. [25] Daytona, “Git Operations | Daytona”, retrieved 2026-10-17. https://www.daytona.io/docs/en/git-operations
  26. [26] Daytona, “Bring Your Own Compute (BYOC) | Daytona”, retrieved 2026-10-17. https://www.daytona.io/docs/en/bring-your-own-compute
  27. [27] Daytona, “daytonaio/daytona (GitHub README)”, retrieved 2026-10-17. https://github.com/daytonaio/daytona
  28. [28] Modal, “Pricing: Pay per second for GPUs and CPUs | Modal”, retrieved 2026-10-17. https://modal.com/pricing
  29. [29] Modal, “Sandboxes (Modal Docs)”, retrieved 2026-10-17. https://modal.com/docs/guide/sandboxes
  30. [30] Modal, “Sandboxes: Isolated environments for reinforcement learning and AI agents | Modal”, retrieved 2026-10-17. https://modal.com/products/sandboxes
  31. [31] Modal, “Networking and security | Modal Docs”, retrieved 2026-10-17. https://modal.com/docs/guide/sandbox-networking
  32. [32] Modal, “Top AI Code Sandbox Products in 2025”, published 2025-07-17. https://modal.com/blog/top-code-agent-sandbox-products
  33. [33] Runloop, “Runloop - Pricing”, retrieved 2026-10-17. https://www.runloop.ai/pricing
  34. [34] Runloop, “Runloop - Your AI Agent Accelerator”, retrieved 2026-10-17. https://www.runloop.ai/
  35. [35] Runloop, “Devbox Overview - Runloop AI Docs”, retrieved 2026-10-17. https://docs.runloop.ai/docs/devboxes/overview
  36. [36] Runloop, “Runloop Features - Runloop AI Docs”, retrieved 2026-10-17. https://docs.runloop.ai/docs/overview/runloop-features
  37. [37] Runloop, “MCP Hub - Runloop AI Docs”, retrieved 2026-10-17. https://docs.runloop.ai/docs/devboxes/mcp-hub
  38. [38] Runloop, “Remote Agents SDK - Runloop AI Docs”, retrieved 2026-10-17. https://docs.runloop.ai/docs/axons/sdk
  39. [39] Vercel, “Vercel Sandbox”, retrieved 2026-10-17. https://vercel.com/docs/sandbox
  40. [40] Vercel, “Vercel Sandbox pricing and quotas”, retrieved 2026-10-17. https://vercel.com/docs/sandbox/pricing
  41. [41] Vercel, “Run untrusted code with Vercel Sandbox, now generally available”, published 2026-01-30. https://vercel.com/blog/vercel-sandbox-is-now-generally-available
  42. [42] Vercel, “Vercel Sandbox vs E2B”, published 2026-03-20. https://vercel.com/kb/guide/vercel-sandbox-vs-e2b
  43. [43] Vercel, “Sandboxes (Vercel MCP tools)”, retrieved 2026-10-17. https://vercel.com/docs/agent-resources/vercel-mcp/tools/sandboxes
  44. [44] Cloudflare, “Sandboxes on Cloudflare”, retrieved 2026-10-17. https://developers.cloudflare.com/sandbox/
  45. [45] Cloudflare, “Choose a sandbox environment”, retrieved 2026-10-17. https://developers.cloudflare.com/sandbox/concepts/
  46. [46] Cloudflare, “Pricing (Containers)”, retrieved 2026-10-17. https://developers.cloudflare.com/containers/platform/pricing/
  47. [47] Cloudflare, “Limits (Containers)”, retrieved 2026-10-17. https://developers.cloudflare.com/containers/platform/limits/
  48. [48] Cloudflare, “Build a coding agent runner”, retrieved 2026-10-17. https://developers.cloudflare.com/sandbox/tutorials/claude-code/
  49. [49] Cloudflare, “Cloudflare's own MCP servers”, retrieved 2026-10-17. https://developers.cloudflare.com/agents/model-context-protocol/cloudflare/servers-for-cloudflare/

Next

Run the next Task on Rush

Connect a repository. Pick Claude Code, Codex, or OpenCode. Review the PR.

How the runtime works