Tensorlake vs Claude Managed Agents
Tensorlake sells sandboxes: microVMs you create through an API and fill with your own agent, with memory snapshots and warm clones. Claude Managed Agents sells the agent itself: you send Claude a task, and Anthropic's harness runs it in a small container it manages.
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
Claude Managed Agents
Claude Managed Agents is Anthropic's ready-made agent harness: Claude runs as an autonomous agent in a cloud sandbox that Anthropic manages, and you talk to it through an events API.
This is an agent platform, not a sandbox you rent. You define an agent (model, prompt, tools, MCP servers, skills) and an environment, then start sessions, send messages as events and stream back what Claude does; the harness handles tool calls, prompt caching and compaction. Each session gets its own Ubuntu container, or runs its tools on a worker you host yourself. It is built for teams that want Claude doing long tasks without building their own agent loop or sandbox.
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 | Claude Managed Agents: price rank | boat: price rank |
|---|---|---|---|
| Software factory 100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month Tensorlake and boat both fit a software factory, and boat ranks 3rd with nested virtualization for emulators; a Managed Agents container with up to 8 GB of memory is a compromise. | 5th cheapest of 367 $1,010 a month | 106th of 367, with limits $964 a month | 3rd cheapest of 367 $634 a month |
| Always-on AI employees 50 machines 24/7 · 2 vCPU / 4 GB Managed Agents lets an idle agent sleep for free and wake on the next message, but its files last only 30 days; Tensorlake keeps a named sandbox's memory while it sleeps. | 9th cheapest of 366 $1,381 a month | 219th of 366, with limits $2,511 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 Tensorlake ranks 24th and fits lots of short runs well; Managed Agents bills runtime only while the agent works, and boat's start-rate limits queue big bursts. | 24th cheapest of 345 $250 a month | 163rd of 345, with limits $756 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 boat ranks 1st and Tensorlake ranks 5th, since both let your users open what runs inside; a Managed Agents sandbox has no inbound ports at all. | 5th cheapest of 361 $471 a month | 231st of 361, with limits $823 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 ranks 1st of 360 for RL rollouts, with up to 1,000 sandboxes at once on Pro and warm clones from one state; Managed Agents is only a compromise here. | Cheapest of 360 $5,637 a month | 252nd of 360, with limits $3,978 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 | 244th of 364, with limits $320 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 boat ranks 4th for CI on its Ryzen 9 9950X hosts; Tensorlake fits too, while Managed Agents documents Docker only as limited availability. | 54th cheapest of 377 $424 a month | 227th of 377, with limits $96 a month | 4th cheapest of 377 $86 a month |
| 24/7 dev boxes 1 machine 24/7 · 4 vCPU / 8 GB A box that stays on all day fits neither cleanly: Tensorlake is built around suspending idle sandboxes and Managed Agents has no SSH, while boat ranks 18th. | 185th of 368, with limits $250 a month | 224th of 368, with limits $48 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 has an XFCE desktop with screenshot, mouse and keyboard calls, Managed Agents has only headless Chromium, and boat streams a full desktop at 60fps. | 9th cheapest of 369 $294 a month | 127th of 369, with limits $288 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 | 13th cheapest of 367 $250 a month | 126th of 367, with limits $288 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 | 127th of 346, with limits $390 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 | 155th of 338, with limits $4,798 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 | 179th of 360, with limits $160 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 | Claude Managed Agents | boat |
|---|---|---|---|
| What it is | A sandbox platform: you create microVMs through an API and run your own agent inside | A hosted Claude agent: Anthropic's harness runs Claude in a sandbox and you send it events | A sandbox platform: each agent gets its own full Linux VM, with coding agents built in |
| Isolation | Firecracker and Cloud Hypervisor microVMs with their own kernel; GPU sandboxes use gVisor | Isolated Linux container per session | Full VM with its own kernel on dedicated AMD Ryzen 9 9950X hosts |
| Machine size | Up to 16 vCPU, 64 GB RAM and 100 GB root disk on Pro, fixed when the sandbox is created | Up to 8 GB memory and 10 GB disk; vCPU not published; not configurable | Up to 16 vCPU / 32 GB on the $100 plan and above |
| Docker and /dev/kvm | Docker on the tensorlake/ubuntu-systemd image; no /dev/kvm (SANDBOX_KVM_AVAILABLE=0) | Docker listed as limited availability; root, systemd and /dev/kvm not documented | Root, Docker, systemd and /dev/kvm, so Android emulators and VMs run |
| Memory snapshots | Yes: suspend and memory snapshots keep running processes and restore warm | No: the sandbox's files and packages are checkpointed when a session goes idle | No: the disk is saved every minute and on stop, and resume reboots the machine |
| Clone or fork | Copy a running or suspended sandbox into many warm clones with memory | No fork or clone of a session or sandbox | Fork a running sandbox with its disk |
| Idle and wake | Idle timeout, 600 s by default, then suspend (named sandboxes) or terminate | An idle session is not billed and resumes from its checkpoint on the next message | Time-to-live (1 h by default, changeable or off); a stopped sandbox costs nothing and is started by API |
| How long state lasts | Named sandboxes keep memory while suspended, billed as snapshot storage | Sandbox state is kept only 30 days after the sandbox is created | The whole disk, for as long as you keep the sandbox |
| Agents | You bring and run your own agent; managed images are plain OS bases | Anthropic's harness with Claude models only, with steering, memory stores, multiagent threads and webhooks | 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 | Vault secrets sit in the sandbox as placeholders and are swapped in at egress, per host | Credentials proxy for model providers and any HTTPS API; real keys never enter the sandbox |
| Inbound traffic | HTTP, WebSocket and gRPC through an authenticated proxy on exposed ports; no raw TCP | No inbound ports or SSH; reached only through the sessions and events API | HTTPS URLs for up to 50 ports, token-protected, plus raw TCP, SSH and a dedicated IP |
| Desktop and browser | XFCE desktop with Firefox over VNC, with screenshot, mouse and keyboard calls | Headless Chromium with Playwright; no desktop or computer-use tool | Full Linux desktop streamed at 60fps, with Chromium, computer use and browser control |
| Outbound rules | allow_out and deny_out lists by domain, IP or CIDR, or no internet at all | Unrestricted, or limited to listed hosts plus optional package registries and MCP servers | Open internet, no provider-side allowlist; you set your own firewall as root |
| Where it runs | US and EU, or your own cloud account through BYOC | US workspace geo only, or tools on workers you host yourself | EU only: Germany, Finland, France |
Every feature Battleships tracks (43)
| Feature | Tensorlake | Claude Managed Agents | boat |
|---|---|---|---|
| Full VM (own kernel) | Yes | No | Yes |
| Docker inside | Yes | Not published | Yes |
| Nested virtualization | No | No | Yes |
| Root | Yes | Not published | Yes |
| Systemd | Yes | Not published | Yes |
| Desktop GUI | Yes | No | Yes |
| Browser | Yes | Yes | Yes |
| Computer-use API | Yes | No | Yes |
| SSH | Yes | No | Yes |
| Persistent disk | Yes | Yes | Yes |
| Pause/resume | Yes | Yes | Yes |
| Automatic snapshots | Yes | Yes | 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 | Yes | 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 | Yes | Yes |
| Secret proxy for any API | No | Yes | Yes |
| Preinstalled agents | No | Yes | Yes |
| Agent harness API | No | Yes | Yes |
| MCP server | No | Yes | No |
| GPU | Yes | No | No |
| Windows | No | No | No |
| Arm64 | No | No | No |
| Self-hosting | No | Yes | 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 | Yes | Yes |
| Audit logs | Not published | Not published | No |
| SSO | Yes | Yes | No |
| SOC 2 | Yes | Yes | No |
| HIPAA | Yes | No | No |
| EU data residency | Yes | No | Yes |
| Zero data retention | Yes | No | Yes |
Performance
There is no comparable CPU measurement for either one: neither Tensorlake nor Claude Managed Agents was part of the hpc-sandbox-benchmarks run, and Anthropic does not publish how many vCPUs a sandbox gets. boat did 26.69 runs per second on the Node.js build-and-test loop at 4 vCPU / 8 GB, the fastest result measured. Tensorlake's docs say its default image starts in a few hundred milliseconds.
Node.js build-and-test loop at 4 vCPU / 8 GB, higher is better. Measured with the open hpc-sandbox-benchmarks suite. Tensorlake and Claude Managed Agents 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 | Claude Managed Agents, 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 | $13 · Claude API (pay-as-you-go, usage tiers) · with limits | $20 · $20/mo |
| One agent that never stops 4 vCPU / 8 GB, 50 GB disk, running 24/7 all month | $250 · Pro | $53 · Claude API (pay-as-you-go, usage tiers) · 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 | $634 · Claude API (pay-as-you-go, usage tiers) · 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 | $360 · Claude API (pay-as-you-go, usage tiers) · 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 | $796 · Claude API (pay-as-you-go, usage tiers) · with limits | $2,000 · $2000/mo · with limits |
Tensorlake: Pro is $250 per billing cycle, which includes $250 of usage, then $0.01 per CU. CPU is billed only while it is actively working, RAM and disk on what you allocate while running, and a suspended sandbox pays snapshot storage.
Tensorlake: Small users can buy prepaid Usage Credits packs from $5 at higher rates instead; sandboxes stop when the balance hits zero. The table prices the Pro path, so it overstates very small workloads.
Claude Managed Agents: Managed Agents bills model tokens at API list rates plus a flat session runtime fee, metered only while the session is running, including time spent waiting on the model. The table shows runtime only; tokens come on top and usually cost more.
Claude Managed Agents: Plans start at $20 a month, and the whole fee is usage credit. You pay per second while a sandbox runs, and a stopped sandbox costs nothing and keeps its disk.
Limits in those scenarios
No scenario counts as a clean fit for Managed Agents: its sandbox has up to 8 GB of memory and 10 GB of disk, an unpublished vCPU count and Docker only in limited availability, and Anthropic publishes no concurrent-session cap, only 300 creates per minute and usage-tier spend limits. Tensorlake fits all five scenarios on Pro, which allows 1,000 sandboxes at once, though an ephemeral sandbox ends after 10 idle minutes by default. boat fits four; at 1,000 short jobs at once even the $2,000 plan queues starts above 90 per minute.
| Scenario | Tensorlake | Claude Managed Agents | 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 run your own agent across many short sandboxes at once, need memory snapshots and warm clones, or want GPUs, US regions or your own cloud account.
- Memory snapshots and warm clones. Useful when agents must resume mid-process: Tensorlake saves a sandbox's memory and copies it into many warm clones.
- Big fan-outs of any agent. Teams running RL rollouts or evals with their own agent can run up to 1,000 sandboxes at once on Pro.
- GPUs, US and EU, or your cloud. Tensorlake offers GPU sandboxes by request and runs in the US, the EU or your own AWS, GCP or Azure account.
Use Claude Managed Agents when
you want Claude itself to do long, mostly text and code tasks with no harness to build, idle time that costs nothing, and the option to run tools on your own workers.
- No agent loop to build. Teams that want Claude doing long tasks get the harness, caching, compaction, memory and webhooks ready to use.
- Idle sessions sleep for free. Agents that wait a long time between messages pay runtime only while working, and resume on the next message.
- Tools on your own workers. Teams with strict network rules can limit sandboxes to listed hosts, or run every tool call on their own infrastructure.
Where boat fits
If your agents do real work for hours, like installs, builds, tests, Docker and emulators, look at boat. It gives each agent a full VM that you can open from outside, with Claude Code, Codex and other agents built in and keys kept outside the machine.
A whole VM with /dev/kvm
Tensorlake has no /dev/kvm, so an Android emulator or a VM does not start. A Managed Agents container has up to 8 GB of memory and Docker only in limited availability, so bigger builds stall.
Open what the agent builds
Managed Agents has no inbound ports or SSH, so you copy the code elsewhere to try it. Tensorlake's proxy carries only HTTP, WebSocket and gRPC, so a raw TCP service like a database can't be reached.
Any agent, keys kept outside
On Tensorlake you install the agent harness yourself and secrets are injected into the sandbox. Managed Agents runs only Claude in Anthropic's harness, so trying Codex on the same task happens somewhere else.
Use boat when your agents need a real machine: Docker, emulators, a desktop, ports you can open, a disk that lasts, and the choice of Claude Code, Codex or other agents. Compare it one to one: boat vs Tensorlake and boat vs Claude Managed Agents.
Questions people ask
Is Tensorlake cheaper than Claude Managed Agents?
For runtime alone, Managed Agents costs less in four of the five scenarios we price, for example $634 against $1,030 for 50 agents in parallel, while Tensorlake costs less for sandboxes for your users ($293 against $360). But Managed Agents adds model tokens on top, which usually cost more than the runtime, and it is not a clean fit for any scenario.
Which is faster, Tensorlake or Claude Managed Agents?
Neither has a result on the hpc-sandbox-benchmarks CPU test, so there is no fair comparison. Tensorlake says its default image starts in a few hundred milliseconds, and Anthropic does not publish its sandbox vCPU count. boat scores 26.69 runs per second on that test.
Can Tensorlake and Claude Managed Agents run Docker?
Tensorlake runs Docker on its tensorlake/ubuntu-systemd image, but has no /dev/kvm. Managed Agents lists Docker as limited availability. boat gives root, Docker, systemd and /dev/kvm in every VM.
Can I run Claude Code on either one?
Managed Agents runs Anthropic's own harness with Claude models, not Claude Code. Tensorlake lets you bring and run your own agent, so you install and wire it up yourself. boat ships Claude Code, Codex and other agents preinstalled.
Which one keeps an agent's state between sessions?
Tensorlake's named sandboxes suspend with memory and running processes. Managed Agents checkpoints files when a session goes idle, but keeps them only 30 days after the sandbox is created. boat keeps the whole disk, saved every minute, until you delete the sandbox.
Is boat SOC 2 compliant?
SOC 2 is in progress. We're open about where we stand: live status at status.ascii.dev and our data retention policy at docs.boat.dev/data-retention.
Try boat
Install the CLI, then run boat new.
7-day free trial