Tensorlake vs AWS AgentCore
Tensorlake sells sandboxes: microVMs that keep their memory while they sleep, copy into warm clones and run with no session cap on Pro. AgentCore is the part of AWS that hosts agents, with 2 vCPU sessions that stop after 8 hours and live inside your AWS account, IAM and VPC.
The companies and products
Tensorlake
Tensorlake is a San Francisco company that sells infrastructure for AI agents, with sandboxes as its main product.
Tensorlake Sandboxes are microVMs, backed by Firecracker and Cloud Hypervisor, that you create through an API and that boot in a few hundred milliseconds. Named sandboxes suspend when idle and keep their memory and running processes, and you can copy a running sandbox into several warm clones. Around them Tensorlake sells versioned file systems, hosted Git, an orchestration layer and a bring-your-own-cloud option, aimed at teams running coding agents, evals and RL rollouts.
- Headquarters
- San Francisco
- Investors shown
- Redpoint and Amplify Partners
AWS AgentCore
Amazon Bedrock AgentCore is a set of AWS services to deploy and run AI agents: a Runtime that hosts your agent code, plus built-in Code Interpreter and Browser tools.
AgentCore is not a sandbox product first. It is a hyperscaler's agent platform: you push a container image or code, and each Runtime session gets its own Firecracker microVM that you reach through AgentCore APIs. Around it sit managed tools (Code Interpreter, Browser) and services for memory, identity, gateways and evaluations, and an Instances compute type runs longer sessions on EC2 in your own account. It is built for teams who already build on AWS and want agents to stay inside their account, IAM and VPC.
Which use cases each one fits
For each use case, Battleships prices every sandbox provider on the same workload and sorts them from the cheapest monthly bill up. “5th cheapest of 367” means 4 of the other 366 providers run that workload, with every feature it needs, for less. “With limits” means the provider runs it but misses something, so it is listed after every full match. Open a use case for its whole ranking and what each provider misses.
| Use case and its workload | Tensorlake: price rank | AWS AgentCore: price rank | boat: price rank |
|---|---|---|---|
| Software factory 100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month Tensorlake fits a software factory with Docker and sessions that can run on; AgentCore is a compromise, with 2 vCPU sessions that stop after 8 hours and keep only a storage mount, and boat adds /dev/kvm and built-in coding agents. | 5th cheapest of 367 $1,010 a month | 100th of 367, with limits $888 a month | 3rd cheapest of 367 $634 a month |
| Always-on AI employees 50 machines 24/7 · 2 vCPU / 4 GB Tensorlake keeps an always-on agent's files and even its memory while it sleeps; on AgentCore it needs the Instances type on EC2 in your account, and boat simply keeps the machine running with no session cap. | 9th cheapest of 366 $1,381 a month | 64th of 366, with limits $1,300 a month | 5th cheapest of 366 $657 a month |
| Code interpreter bursts 100 at once · 1 vCPU / 2 GB · 200,000 sessions of 1 min a month AgentCore Code Interpreter is built for short code runs billed only for CPU used, so it fits this best of the three; Tensorlake fits too, and boat's start-rate limits make bursts a compromise. | 24th cheapest of 345 $250 a month | 7th cheapest of 345 $114 a month | 186th of 345, with limits $2,000 a month |
| Sandboxes for your users 300 at once · 2 vCPU / 4 GB · 30,000 sessions of 23 min a month Tensorlake gives each user a sandbox with HTTP, WebSocket and gRPC through its proxy, while AgentCore sessions have no open ports; boat adds HTTPS URLs on up to 50 ports and raw TCP. | 5th cheapest of 361 $471 a month | 210th of 361, with limits $506 a month | Cheapest of 361 $209 a month |
| RL rollouts 1,000 at once · 2 vCPU / 4 GB · 1,000,000 sessions of 3 min a month Tensorlake is built for RL rollouts, with warm clones from one prepared state and 1,000 sandboxes at once on Pro; AgentCore has no snapshots or fork, and boat's start-rate limits make it a compromise. | Cheapest of 360 $5,637 a month | 125th of 360, with limits $8,075 a month | 34th of 360, with limits $2,000 a month |
| Agent evals 250 at once · 4 vCPU / 8 GB · 15,000 sessions of 20 min a month | 24th cheapest of 364 $506 a month | 33rd cheapest of 364 $650 a month | 10th cheapest of 364 $200 a month |
| CI runners 40 at once · 8 vCPU / 16 GB · 6,000 sessions of 12 min a month | 54th cheapest of 377 $424 a month | 52nd cheapest of 377 $399 a month | 4th cheapest of 377 $86 a month |
| 24/7 dev boxes 1 machine 24/7 · 4 vCPU / 8 GB Neither fits a box that stays on all day cleanly: Tensorlake's model is built around suspending idle sandboxes, and AgentCore has no SSH and resets files outside its mount; boat keeps the machine on and its whole disk. | 185th of 368, with limits $250 a month | 228th of 368, with limits $52 a month | 18th cheapest of 368 $26 a month |
| Computer-use agents 50 at once · 4 vCPU / 8 GB · 20,000 sessions of 15 min a month Tensorlake gives agents an XFCE desktop with Firefox and mouse and keyboard calls, while AgentCore Browser drives only a browser; boat streams a full Linux desktop at 60fps. | 9th cheapest of 369 $294 a month | 104th of 369, with limits $202 a month | 6th cheapest of 369 $200 a month |
| Browser-use agents 50 at once · 2 vCPU / 4 GB · 30,000 sessions of 10 min a month AgentCore Browser is a managed Chromium with live view and session replay, Tensorlake runs a browser on its desktop image, and boat runs Chromium on a full machine you control. | 13th cheapest of 367 $250 a month | 4th cheapest of 367 $144 a month | 7th cheapest of 367 $200 a month |
| Scraping & bots (startup scale) 50 at once · 1 vCPU / 2 GB · 100,000 sessions of 3 min a month | 22nd cheapest of 346 $250 a month | 27th cheapest of 346 $287 a month | 6th cheapest of 346 $90 a month |
| Mass scraping & bots (undetectable) 1,000 at once · 1 vCPU / 2 GB · 3,000,000 sessions of 2 min a month | 121st of 338, with limits $3,484 a month | 165th of 338, with limits $5,458 a month | 71st of 338, with limits $1,800 a month |
| Parallel red teaming 200 at once · 2 vCPU / 4 GB · 5,000 sessions of 30 min a month | 4th cheapest of 360 $250 a month | 62nd of 360, with limits $294 a month | 2nd cheapest of 360 $200 a month |
| GPU eval jobs 10 at once · 8 vCPU / 32 GB · 500 sessions of 30 min a month | Can't run it | Can't run it | Can't run it |
| macOS agents 5 at once · 4 vCPU / 8 GB · 1,000 sessions of 20 min a month | Can't run it | Can't run it | Can't run it |
Key features
| Feature | Tensorlake | AWS AgentCore | boat |
|---|---|---|---|
| What it is | A sandbox product: microVMs you create through an API | An AWS agent platform: hosted agent sessions plus managed Code Interpreter and Browser tools | A sandbox platform: each agent gets its own full Linux machine |
| Isolation | Firecracker and Cloud Hypervisor microVMs with their own kernel; GPU sandboxes use gVisor | Firecracker microVM per Runtime session; Code Interpreter and Browser in a containerized environment | Full VM with its own kernel on dedicated AMD Ryzen 9 9950X hosts |
| Biggest machine | Up to 16 vCPU / 64 GB with a 100 GB root disk on Pro | 2 vCPU / 8 GB per microVM session, not adjustable; bigger EC2 types only on Instances | Up to 16 vCPU / 32 GB on the $100 plan and above |
| Session length | No cap on Pro; an idle timeout (600 s by default) suspends or ends the sandbox | Up to 8 hours on microVMs, stopped after 15 idle minutes by default; up to 14 days on Instances | No cap on paid plans; you set the time to live, with no idle timer |
| What survives a stop | Named sandboxes suspend with their disk, memory and running processes | Only one storage mount under /mnt, up to 1 GB, in preview; EBS volumes on Instances; the rest resets | The whole disk, saved every minute and on stop; a stopped sandbox costs nothing |
| Snapshots and clones | Memory snapshots; copy a running sandbox into many warm clones | No user snapshots or fork of a session | Automatic disk snapshots every minute; fork a running sandbox with its disk |
| Docker, systemd and /dev/kvm | Docker and systemd on the tensorlake/ubuntu-systemd image; no /dev/kvm | Root inside the microVM with arm64 container images; Docker and /dev/kvm not documented | Root, Docker, systemd and /dev/kvm, so Android emulators and VMs run |
| Inbound access | HTTP, WebSocket and gRPC through an authenticated proxy on exposed ports | No open ports or SSH; sessions are reached only through AgentCore APIs with IAM or JWT | HTTPS URLs for up to 50 ports, token-protected, plus raw TCP, SSH and a dedicated IP |
| Network controls | allow_out and deny_out lists by domain, IP or CIDR, or no internet at all | VPC mode with security groups, PrivateLink and a NAT gateway for a fixed outbound IP | Open internet, no provider-side allowlist; you set your own firewall as root |
| Browser and desktop | XFCE desktop with Firefox over VNC, with screenshot, mouse and keyboard calls | Managed Chromium with live view, session recording and replay; no full desktop | Full Linux desktop streamed at 60fps, with Chromium, computer use and browser control |
| Coding agents | You bring and run your own agent; managed images are plain OS bases | AWS's managed harness built on Strands Agents, or your own agent code | Claude Code, Codex, OpenCode and more preinstalled, driven from outside with a prompt, steer and events API |
| API keys | Stored secrets are injected into the sandbox | AgentCore Identity stores OAuth tokens and API keys and hands them to your agent code | Credentials proxy for model providers and any HTTPS API: real keys never enter the sandbox |
| GPUs | GPU sandboxes by request (H100 shown in the docs) | NVIDIA G-series and Inferentia2 on the Instances type | None |
| Where it runs | US and EU, or your own cloud account through BYOC | 17 commercial AWS regions plus GovCloud, or EC2 in your account on Instances | EU only: Germany, Finland, France, with EU data residency |
Every feature Battleships tracks (43)
| Feature | Tensorlake | AWS AgentCore | boat |
|---|---|---|---|
| Full VM (own kernel) | Yes | Yes | Yes |
| Docker inside | Yes | Not published | Yes |
| Nested virtualization | No | No | Yes |
| Root | Yes | Yes | Yes |
| Systemd | Yes | No | Yes |
| Desktop GUI | Yes | No | Yes |
| Browser | Yes | Yes | Yes |
| Computer-use API | Yes | Yes | Yes |
| SSH | Yes | No | Yes |
| Persistent disk | Yes | Yes | Yes |
| Pause/resume | Yes | No | Yes |
| Automatic snapshots | Yes | No | Yes |
| Fork/clone | Yes | No | Yes |
| Live fork (no pause) | Yes | No | Yes |
| Memory fork | Yes | No | No |
| Custom image (Docker or snapshot) | Yes | Yes | Yes |
| ≥24 h sessions | Yes | No | Yes |
| Idle auto-stop | Yes | Yes | No |
| Wake on request | Yes | Yes | No |
| HTTPS preview URLs | Yes | No | Yes |
| Raw TCP inbound | No | No | Yes |
| Public IPv4 | No | No | No |
| Egress allowlist | Yes | Yes | No |
| Secret proxy | No | No | Yes |
| Secret proxy for any API | No | No | Yes |
| Preinstalled agents | No | No | Yes |
| Agent harness API | No | No | Yes |
| MCP server | No | Yes | No |
| GPU | Yes | Yes | No |
| Windows | No | No | No |
| Arm64 | No | Yes | No |
| Self-hosting | No | No | No |
| BYOC | Yes | Yes | No |
| Open source | No | No | No |
| Scoped API keys | Not published | Not published | Yes |
| Spend limits | Not published | Yes | Yes |
| Usage per sandbox (API) | No | No | Yes |
| Audit logs | Not published | Yes | No |
| SSO | Yes | Yes | No |
| SOC 2 | Yes | Yes | No |
| HIPAA | Yes | Yes | No |
| EU data residency | Yes | Yes | Yes |
| Zero data retention | Yes | Not published | Yes |
Performance
There is no comparable CPU measurement for Tensorlake or AWS AgentCore: neither was part of the hpc-sandbox-benchmarks run. boat scores 26.69 runs/s on its Node.js build-and-test loop at 4 vCPU / 8 GB, the fastest result measured. An AgentCore microVM session stops at 2 vCPU, so a like-for-like test would need its Instances type on EC2.
Node.js build-and-test loop at 4 vCPU / 8 GB, higher is better. Measured with the open hpc-sandbox-benchmarks suite. Tensorlake and AWS AgentCore have no published run in this suite yet, so there is no like-for-like number to compare.
Pricing in five scenarios
Each bill is priced from the providers' public price lists by the open Battleships engine, with the cheapest plan that runs the scenario.
| Scenario | Tensorlake, a month | AWS AgentCore, a month | boat, a month |
|---|---|---|---|
| One coding agent, workdays 4 vCPU / 8 GB, 50 GB disk, 8 hours a day, 22 days a month, files kept between days | $250 · Pro | $26 · Pay-as-you-go (Instances) · with limits | $20 · $20/mo |
| One agent that never stops 4 vCPU / 8 GB, 50 GB disk, running 24/7 all month | $250 · Pro | $107 · Pay-as-you-go (Instances) · with limits | $26 · $20/mo |
| 50 agents in parallel 50 machines of 4 vCPU / 8 GB at once, 8-hour sessions, 1,100 sessions a month, Docker inside | $1,030 · Pro | $1,286 · Pay-as-you-go (Instances) · with limits | $317 · $20/mo × 3 seats |
| Sandboxes for your users 2 vCPU / 4 GB, 30,000 sessions of 10 minutes a month, up to 300 at once | $293 · Pro | $423 · Pay-as-you-go · with limits | $200 · $100/mo × 2 seats |
| 1,000 short jobs at once 2 vCPU / 4 GB, 1,000 at once, 3 minutes each, 200,000 jobs a month | $1,127 · Pro | $1,615 · Pay-as-you-go (Instances) | $2,000 · $2000/mo · with limits |
Tensorlake: Pro is $250 per billing cycle, which includes 25,000 CU of usage, then $0.01 per CU. CPU is billed only while it is actively working, RAM and disk on what you allocate while the sandbox runs, and a suspended sandbox pays snapshot storage per GB-month.
Tensorlake: Small users can skip Pro and buy prepaid Usage Credits packs from $5 at higher rates, with no overage: sandboxes stop when the balance hits zero. The table prices the Pro path, so it overstates very small workloads.
AWS AgentCore: Runtime microVMs, Code Interpreter and Browser bill per second for CPU actually used and for peak memory. The 4 vCPU shape in these scenarios is the Instances type: the EC2 on-demand price plus a 12% management fee, with EBS storage and data transfer billed by AWS on top.
AWS AgentCore: Plans start at $20 a month, and the whole fee becomes usage credit. Running time is billed per second, and a stopped sandbox costs nothing while its disk is kept.
Limits in those scenarios
Tensorlake fits all five scenarios on Pro, which allows 1,000 sandboxes at once; its limits are that CPU, memory and disk are fixed at creation, disk is 10 to 100 GiB, and an ephemeral sandbox ends after 10 idle minutes by default. AgentCore's microVM sessions stop at 2 vCPU / 8 GB and 8 hours, so the 4 vCPU scenarios need the Instances type, and files outside the storage mount reset after a stop, which is why it fits only the 1,000 short jobs scenario. boat has no session cap, but its start limits mean 1,000 short jobs at once does not fit even on the $2,000/mo plan.
| Scenario | Tensorlake | AWS AgentCore | boat |
|---|---|---|---|
| One coding agent, workdays | None that bite |
| None that bite |
| One agent that never stops | None that bite |
| None that bite |
| 50 agents in parallel | None that bite |
| None that bite |
| Sandboxes for your users | None that bite |
| None that bite |
| 1,000 short jobs at once | None that bite |
|
|
Which one should you use?
Use Tensorlake when
you fan out many short rollouts from one warm state, need sandboxes that sleep with their memory, or want them in your own cloud account.
- Memory snapshots and warm clones. Useful when agents must pick up mid-process: Tensorlake saves a sandbox's memory and copies it into warm clones.
- Bigger sandboxes with no session cap. For agents that need more than 2 vCPU or run past 8 hours, Tensorlake Pro goes up to 16 vCPU with no session cap.
- Reach the app inside. Tensorlake exposes HTTP, WebSocket and gRPC ports through its proxy, so you can open what an agent is serving.
Use AWS AgentCore when
your agents must stay inside your AWS account and VPC, need a GPU, or are short code and browser tasks that fit the managed tools.
- It lives inside your AWS account. Teams built on AWS keep agents behind their own IAM, VPC, security groups and PrivateLink, in 17 commercial regions.
- Managed tools for short tasks. For short code runs and browser tasks, Code Interpreter and Browser charge only for CPU used, with live view and replay.
- GPUs and two-week sessions on Instances. Agents that need a GPU or a session of up to 14 days can run on EC2 in your account, with Savings Plans applying.
Where boat fits
If your agents do long, real work on a machine, such as installs, builds, tests, Docker, emulators and previews, look at boat. It gives each agent a full VM on dedicated AMD Ryzen 9 9950X hosts that keeps its whole disk, runs with no session cap on paid plans and can be reached like a server. Tensorlake is still ahead for warm memory clones and GPUs, and AgentCore for staying inside AWS.
Run emulators and VMs inside
Tensorlake sandboxes have no /dev/kvm, and AgentCore does not document Docker or /dev/kvm in a session, so an Android emulator or a VM an agent needs to test with does not start. boat exposes /dev/kvm with Docker and systemd.
Open any port the agent starts
Tensorlake's proxy carries only HTTP, WebSocket and gRPC, so a database or raw TCP service can't be reached, and AgentCore has no open ports at all, so you must proxy traffic through your own code. boat gives HTTPS on 50 ports and raw TCP.
Coding agents built in, keys kept outside
On both, you install the agent harness yourself or use AWS's own, and API keys end up in the sandbox or in your agent code, where one prompt injection can leak them. boat preinstalls Claude Code and Codex and keeps real keys in a proxy outside.
Use boat when your agents do real work for hours: installs, builds, tests, Docker, emulators and a desktop, with a disk that keeps everything, ports you can open and keys kept outside the machine. Compare it one to one: boat vs Tensorlake and boat vs AWS AgentCore.
Questions people ask
Is Tensorlake cheaper than AWS AgentCore?
For 1,000 short jobs at once, the one scenario both fit, Tensorlake is $1,127 a month against $1,615 on AgentCore. AgentCore looks lower for one coding agent on workdays ($26 against $250), but it does not fit, because installed packages reset after each stop. For short code runs, AgentCore Code Interpreter is the cheaper fit.
Which is faster, Tensorlake or AgentCore?
Neither has a published result on the hpc-sandbox-benchmarks CPU test, so there is no fair speed comparison. On size, Tensorlake Pro goes up to 16 vCPU, while an AgentCore microVM session stops at 2 vCPU. boat scores 26.69 runs/s, the fastest measured.
Can Tensorlake or AgentCore run Docker?
Tensorlake runs Docker on its tensorlake/ubuntu-systemd image. AgentCore runs your arm64 container image as root in a microVM, but its docs do not say Docker works inside a session. boat runs Docker with root and systemd.
How long can a session run on each?
Tensorlake Pro has no session cap, though an idle timeout of 600 seconds by default suspends or ends a sandbox. An AgentCore microVM session lasts up to 8 hours and stops after 15 idle minutes by default; the Instances type runs up to 14 days. boat has no session cap on paid plans.
Which one keeps a sandbox's state?
Tensorlake named sandboxes suspend with their memory and running processes, and can be copied into warm clones. AgentCore keeps only a storage mount of up to 1 GB, in preview, or EBS volumes on Instances. boat keeps the whole disk, saved every minute, and can fork a running sandbox.
Do they support nested virtualization?
Tensorlake sandboxes set SANDBOX_KVM_AVAILABLE=0, so no. AgentCore does not document /dev/kvm. boat exposes /dev/kvm, so Android emulators and VMs run inside.
Try boat
Install the CLI, then run boat new.
7-day free trial