AWS Lambda MicroVMs
Isolated sandboxes. Near-instant launch and resume. Full lifecycle control.
What is AWS Lambda MicroVMs?
Lambda MicroVMs is a serverless compute primitive that combines VM-level isolation, near-instant launch and resume, and state persistence across interactions. Built on the same Firecracker virtualization powering 15 trillion+ monthly Lambda invocations, MicroVMs are purpose-built for workloads that execute user- or AI-generated code.
Start a MicroVM, connect over HTTP, and begin executing code for your AI or human users. No infrastructure to manage, no capacity to plan, no isolation expertise required.
What can you build with Lambda MicroVMs?
Isolated, stateful code execution at Lambda speed
Launch your own isolated Firecracker-powered sandbox that starts near-instantly and preserves full memory and disk state for up to 8 hours. Suspend automatically when idle, resume on demand. Built on the same virtualization technology powering 15 trillion monthly Lambda requests.
Build multi-tenant workloads with serverless simplicity
Launch each MicroVM with a dedicated URL and JWE-based authentication. Simply send each tenant's traffic to their assigned MicroVM. Lambda manages everything from network routing to isolation. Run each tenant on an isolated kernel, memory space, and disk state.
Flexible environment scaling, lifecycle, and state retention
Whether your user sessions last for seconds or several hours, scale resource consumption on demand. MicroVMs can vertically scale up to 4x your configured baseline resources during peak periods. Control the full lifecycle: launch, suspend, resume, and terminate on your terms.
Where does Lambda MicroVMs fit in your architecture?
Lambda MicroVMs is purpose-built for workloads that need isolation, state, and lifecycle control.
AI coding assistants and agent sandboxes
AI-powered development tools generate and execute code on behalf of users. If this code runs inside the agent's runtime, it can access credentials and memory. Lambda MicroVMs provides a separate execution boundary per task: an isolated compute environment with no access to agent state and no shared state across users. State persists across multi-step tasks. No infrastructure to manage.
Interactive development platforms
Browser-based development platforms need each user to get their own isolated environment instantly. Lambda MicroVMs runs every session in an isolated compute environment with packages and dependencies preserved across interactions, without security hardening complexity.
Vulnerability scanning and security testing
Security tools executing user- or AI-generated code for analysis require VM-level containment. Lambda MicroVMs provides an isolated compute environment per session, preventing malicious code from affecting other users or the host. Environments spin up and tear down on demand, with no residual state between scans.
Data analytics platforms
Data scientists need environments that preserve packages, results, and execution context across sessions without resetting. Interactive query engines need rapid startup with per-tenant isolation. Lambda MicroVMs meets the needs of both workloads: each user gets an isolated compute environment that retains full state for up to 8 hours, suspends when idle, and resumes near-instantly. No re-computation between sessions.
CI/CD and developer productivity platforms
CI/CD platforms routinely run user- or AI-generated code from pull requests and third-party dependencies. Lambda MicroVMs provides an isolated compute environment per pipeline stage. Environments launch near-instantly and resume where they left off. No shared state between jobs.
Caylent accelerates agentic workloads with isolated, stateful compute
Caylent partners with enterprises to design, build, and ship agentic products, platforms, and systems on AWS. The company has helped hundreds of organizations take AI agents from prototype to production, and as those workloads shift toward long-running, stateful execution, the underlying compute layer becomes a critical bottleneck. Lambda MicroVM removes that constraint by providing a lightweight, isolated execution environment purpose-built for the agentic patterns customers need in production. By building on Lambda MicroVM, organizations move faster without the infrastructure overhead that slows most teams down.
"AWS continues to broaden the available compute options with Lambda MicroVM and this is one of the most interesting new AWS primitives we have tested for stateful, agentic workloads. It is one of the best primitives to run your AI agents, developer sandboxes, data-processing workers, and interactive automation systems that need to preserve execution state across pauses and resumes. For teams already building on AWS, Lambda MicroVM offers a simpler path to isolated, long-running execution environments without taking on the full operational burden of managing clusters, capacity, host maintenance, networking, and VM lifecycle automation."
—Randall Hunt, CTO at Caylent
ClickHouse delivers a private SQL engine in every isolated workload
“At ClickHouse, we build chDB, an in-process ClickHouse you get with a single ‘pip install chdb’, giving any application a full SQL engine, federated access to 80+ formats and to S3, Postgres, and remote ClickHouse, and vector + time-series memory, all with zero network round-trips. AWS Lambda MicroVMs is a perfect match for that model: Firecracker hardware isolation, snapshot-based fast starts, and suspend/resume let every isolated workload carry its own private chDB engine, hot from the first millisecond and billed only while it runs. Because chDB pushes compute down to the data and returns only small answers, it’s an ideal fit for the MicroVM’s dedicated endpoint. We’re building a set of reference architectures on this combination: a federated query hub that joins local data, S3, CDNs, and Postgres in one query inside a single MicroVM; isolated CI/CD runners where every test gets a clean chDB and there’s no shared server to overload; on-demand sandboxes where an AI agent can spend hours reproducing a bug in a disposable VM; and a per-session ‘agent brain’ where chDB is the local memory and federation layer the agent thinks with, suspended and resumed right alongside the user’s session. ClickHouse Cloud is where your data lives, chDB is what your agent thinks with, and Lambda MicroVMs is where it gets to think, in private.”
—Auxten Wang, Director, ClickHouse (creator of chDB)
Siemens delivers isolated agentic coding sandboxes to developers in hours
"Setting up Lambda MicroVMs was remarkably straightforward. Within hours, we were able to offer our developers on our team a fully isolated, secured sandbox environment with agentic coding capabilities that spins up in seconds. It's the speed of serverless with the isolation of a dedicated VMs, exactly what we needed to accelerate our development cycles."
—Christoph Lumme, Enterprise Architect, Siemens Switzerland Ltd
Snorkel AI scales agent simulations with secure, cost-efficient execution
Snorkel AI builds frontier datasets for agentic systems, which requires operating across the full AI lifecycle - from curated datasets like Snorkel Data Series to evaluation, reinforcement learning, and large-scale agent simulation. The team runs millions of agent simulations each month to ensure its environments stay challenging for frontier models while remaining grounded in real-world tasks and constraints. Lambda MicroVMs are a strong fit for these workloads because they can provide secure, tightly permissioned execution environments for heterogeneous simulation and evaluation jobs. For agent workflows that vary widely in runtime and resource demand, that combination of isolation and usage-based CPU and memory pricing could make it much easier to scale safely and efficiently.
“Snorkel runs large volumes of agent simulations and evaluation workloads, so we care deeply about secure isolation, tight permissions, and efficient scaling. Lambda MicroVMs are compelling because they could let us run diverse agent workflows in sandboxed environments while aligning cost much more closely to the CPU and memory each job actually uses.”
—Rustem Feyzkhanov, Sr. Manager, AI Platform, Snorkel AI
How much does Lambda MicroVMs cost?
Did you find what you were looking for today?
Let us know so we can improve the quality of the content on our pages