Does E2B run my LLM, or just the code?

Just the code and tools — you connect your own LLM. E2B is model-agnostic sandbox infrastructure.

Can a sandbox keep state between runs?

Yes. Pausing saves both the filesystem and memory, and paused sandboxes are kept indefinitely until you resume them.

HostingOpen sourceNot yet hands-on tested

E2B

An open-source, secure environment with real-world tools for enterprise-grade agents.

In brief

E2B is an agent hosting & compute: An open-source, secure environment with real-world tools for enterprise-grade agents. Hosting is hosted cloud that spins up on-demand secure linux vm sandboxes reachable both internally and via a sandbox url; a free tier is available to start; and it's native on MCP. Best suited to teams that want to run their own agent on managed compute.

What it is

E2B provides isolated, secure Linux sandboxes created on demand so AI agents can execute code, process data, and run automations. It is infrastructure — an SDK plus hosted cloud — rather than an agent framework or a model.

At a glance

7/7 facts sourced

Pricing: Hobby (free) includes a one-time $100 usage credit, up to 1-hour sessions, and 20 concurrent sandboxes. Pro is $150/mo + usage (24-hour sessions, up to 100 concurrent). Usage is per-second (e.g. 1 vCPU at $0.000014/s).source ↗
Hosting model: Hosted cloud that spins up on-demand secure Linux VM sandboxes reachable both internally and via a sandbox URL; self-hostable given the open-source codebase.source ↗
Memory & state: Sandbox persistence — pausing saves both filesystem and memory state, and connect() resumes the identical state, with paused sandboxes kept indefinitely. This is infrastructure state, not agent/LLM memory.source ↗
MCP support: Nativesource ↗
Integrations: First-party Python and JS/TS SDKs, Templates for defining environments, and MCP tools from a Docker catalog (Browserbase, Stripe, GitHub, Notion, and more).source ↗
Model providers: Model-agnostic — you bring your own LLM; E2B runs the code and tools, not the model.source ↗
Open source: Yessource ↗

3.7of 5

Editorial scorecard

Overall 3.7 / 5

Our opinion — the same 0–5 rubric is applied to every platform, including Alfe. See the rubric.

Developer experience4/5

Pricing transparency & value4/5

Scalability4/5

Memory & state3/5

Integrations3/5

MCP nativeness4/5

Strengths

Pause/resume preserves both filesystem and memory, retained indefinitely.
Free Hobby tier with a $100 credit and 20 concurrent sandboxes; transparent per-second pricing.
Open source with first-party SDKs and an official MCP server.

Trade-offs

Infrastructure only — no LLMs, orchestration, or agent memory; you bring your own.
A single continuous session is capped (1 hour on Hobby, 24 hours on Pro) before you must pause and resume.
Usage-based per-second cost plus the $150/mo Pro base is harder to forecast than flat pricing.

Our take

Editorial opinion — not a hands-on test result

In our opinion, E2B is a strong pick when you specifically need secure, disposable sandboxes for agent-run code and want to bring your own model and orchestration. It is not a full agent platform, so treat it as a building block. The pause/resume persistence is a genuine differentiator for long-running or resumable agent tasks.

FAQ

Does E2B run my LLM, or just the code?: Just the code and tools — you connect your own LLM. E2B is model-agnostic sandbox infrastructure.
Can a sandbox keep state between runs?: Yes. Pausing saves both the filesystem and memory, and paused sandboxes are kept indefinitely until you resume them.

Compare E2B

Sources

E2B homepage · accessed 2026-07-01
E2B pricing · accessed 2026-07-01
E2B sandbox persistence docs · accessed 2026-07-01
E2B on GitHub · accessed 2026-07-01

The Agent Examiner editorial · last updated 2026-07-01. Facts are drawn from each vendor’s own public materials; scores are our editorial opinion (methodology).

Product and company names are trademarks of their respective owners and are used here only to identify the products under discussion.