Tensorlake vs Together Code Sandbox
Both run each sandbox in its own VM and can save its memory. Tensorlake is self-serve and can copy a running sandbox into warm clones, while Together's VM product must be enabled for your organization and saves a snapshot only when a sandbox ends.
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
Together Code Sandbox
Together AI is a cloud for running and training AI models, and Code Sandbox is its product for running code in isolated VMs.
You build a snapshot from a Docker image or Dockerfile on Together's builder, then start sandboxes from it and run commands and file operations through the together-sandbox SDK or CLI. When a sandbox ends you can save its disk, and its memory too, as a new snapshot to start more sandboxes from. It is aimed at model-generated code, agent tool calls and reinforcement learning rollouts, and it must be enabled for your organization by Together. Together also sells Code Interpreter, a separate API that runs Python code in short sessions.
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 | Together Code Sandbox: price rank | bochka: price rank |
|---|---|---|---|
| Software factory 100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month Tensorlake ranks 5th and fits a software factory, while the self-serve Together option is a Python runner that ranks 65th; bochka ranks 3rd and adds /dev/kvm. | 5th cheapest of 367 $1,010 a month | 65th of 367, with limits $528 a month | 3rd cheapest of 367 $634 a month |
| Always-on AI employees 50 machines 24/7 · 2 vCPU / 4 GB Tensorlake keeps an agent's memory while it sleeps and wakes on request; Together sandboxes have no idle detection or wake, and bochka keeps the machine on with no session cap. | 9th cheapest of 366 $1,381 a month | 152nd of 366, with limits $1,095 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 for short code runs; Together's Code Interpreter suits plain Python in 60-minute sessions but keeps nothing after a session ends. | 24th cheapest of 345 $250 a month | 326th of 345, with limits $2,190 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 routes each user's app through an authenticated proxy, Together leaves every port public, and bochka ranks 1st with token-protected HTTPS URLs and raw TCP. | 5th cheapest of 361 $471 a month | 234th of 361, with limits $900 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 for RL rollouts, with up to 1,000 sandboxes at once on Pro and warm clones from one running state. | Cheapest of 360 $5,637 a month | 315th of 360, with limits $21,900 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 | 258th of 364, with limits $450 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 | 251st of 377, with limits $180 a month | 4th cheapest of 377 $86 a month |
| 24/7 dev boxes 1 machine 24/7 · 4 vCPU / 8 GB Neither is built for a box that stays on all day: Tensorlake suspends idle sandboxes and Together's self-serve Code Interpreter sessions end after an hour, while bochka ranks 18th and fits. | 185th of 368, with limits $250 a month | 210th of 368, with limits $22 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 Firefox over VNC, Together has no desktop at all, and bochka streams a 60fps desktop with Chromium. | 9th cheapest of 369 $294 a month | 184th of 369, with limits $600 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 | 301st of 367, with limits $900 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 | 219th of 346, with limits $180 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 | 124th of 338, with limits $3,600 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 Tensorlake gives root in a real kernel with outbound allow and deny lists; bochka ranks 2nd and adds raw TCP inbound and nested virtualization. | 4th cheapest of 360 $250 a month | 167th of 360, with limits $150 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 | Together Code Sandbox | bochka |
|---|---|---|---|
| Isolation | Firecracker and Cloud Hypervisor microVMs; GPU sandboxes use gVisor | A VM per sandbox on Together's infrastructure | Full VM with its own kernel on dedicated AMD Ryzen 9 9950X hosts |
| Getting started | Self-serve: Free tier, prepaid credit packs from $5, or Pro | The VM SDK and CLI must be enabled for your organization; only Code Interpreter is self-serve | Sign up and start a sandbox from the API, CLI or dashboard |
| Machine size | Up to 16 vCPU, 64 GB RAM and 100 GB root disk on Pro, fixed at creation | 0.1 to 16 cores, 1 to 8 GB of memory per core; disk size fixed | 2 to 16 vCPU with 2 GB per vCPU, 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 | Commands run as user 1000 by default; Docker inside and nested virtualization not documented | Docker, systemd, root and /dev/kvm on every machine |
| Memory snapshots | Suspend or snapshot a sandbox with its memory and running processes | When a sandbox ends, save its disk, or disk and memory, as a snapshot | Disk saved every minute and on stop; resume boots the machine again |
| Clone a running sandbox | Copy a running or suspended sandbox into many warm clones with memory | No live fork: the parent stops, then you start many sandboxes from its snapshot | Fork a running sandbox (disk); the original keeps running |
| Idle and wake | Idle timeout, 600 s by default, then suspend or terminate | Time to live counts from creation; no idle detection and no wake on request | Time to live, 1 h by default, changeable or off; a stopped sandbox costs nothing |
| Desktop and computer use | XFCE desktop with Firefox over VNC, with screenshot, mouse and keyboard calls | None; headless command and file access only | Linux desktop streamed at 60fps, Chromium and computer-use tools |
| Coding agents | You bring and run your own agent; managed images are plain OS bases | Run your own agent through execs; a Harbor guide for agent evals | Claude Code, Codex, OpenCode and more preinstalled, with a prompt, steer and events API |
| API keys | Stored secrets are injected into the sandbox | No credentials proxy documented | 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 | All ports public; preview URLs coming soon; an experimental inbound allowlist | HTTPS URLs for up to 50 ports, token-protected by default, plus raw TCP |
| Outbound rules | allow_out and deny_out lists by domain, IP or CIDR, or no internet at all | No outbound allowlist documented | No provider-side allowlist; you set your own firewall as root |
| GPUs and regions | GPU sandboxes by request; US and EU, or your own cloud through BYOC | No GPU sandboxes; regions not published | No GPUs; EU only: Germany, Finland, France |
| Hosted Python runner | No separate product; run code inside a sandbox | Code Interpreter API runs Python in sessions billed per 60 minutes | No separate product; run code through exec in a sandbox |
Every feature Battleships tracks (43)
| Feature | Tensorlake | Together Code Sandbox | bochka |
|---|---|---|---|
| Full VM (own kernel) | Yes | Yes | Yes |
| Docker inside | Yes | Not published | Yes |
| Nested virtualization | No | No | Yes |
| Root | Yes | Yes | Yes |
| Systemd | Yes | Not published | Yes |
| Desktop GUI | Yes | No | Yes |
| Browser | Yes | No | Yes |
| Computer-use API | Yes | No | Yes |
| SSH | Yes | No | Yes |
| Persistent disk | Yes | Yes | Yes |
| Pause/resume | Yes | Yes | Yes |
| Automatic snapshots | Yes | No | Yes |
| Fork/clone | Yes | Yes | Yes |
| Live fork (no pause) | Yes | No | Yes |
| Memory fork | Yes | Yes | No |
| Custom image (Docker or snapshot) | Yes | Yes | Yes |
| ≥24 h sessions | Yes | Yes | Yes |
| Idle auto-stop | Yes | No | No |
| Wake on request | Yes | No | No |
| HTTPS preview URLs | Yes | Yes | Yes |
| Raw TCP inbound | No | No | Yes |
| Public IPv4 | No | No | No |
| Egress allowlist | Yes | No | 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 | No | No |
| Windows | No | No | No |
| Arm64 | No | No | No |
| Self-hosting | No | No | No |
| BYOC | Yes | No | 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 | Not published | Yes |
| Zero data retention | Yes | Not published | Yes |
Performance
Neither Tensorlake nor Together Code Sandbox has a result on the StarSling hpc-sandbox-benchmarks CPU test, so there is no comparable CPU measurement between them. bochka did 26.69 runs per second on its Node.js test at 4 vCPU / 8 GB, on dedicated AMD Ryzen 9 9950X hosts.
Node.js build-and-test loop at 4 vCPU / 8 GB, higher is better. Measured with the open hpc-sandbox-benchmarks suite. Tensorlake and Together Code Sandbox 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 | Together Code Sandbox, a month | bochka, 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 | $5 · Together pay-as-you-go (prepaid credits; Code Interpreter only) · with limits | $20 · $20/mo |
| One agent that never stops 4 vCPU / 8 GB, 50 GB disk, running 24/7 all month | $250 · Pro | $22 · Together pay-as-you-go (prepaid credits; Code Interpreter only) · 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 | $264 · Together pay-as-you-go (prepaid credits; Code Interpreter only) · 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 | $900 · Together pay-as-you-go (prepaid credits; Code Interpreter only) · 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 | $6,000 · Together pay-as-you-go (prepaid credits; Code Interpreter only) · with limits | $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 active, RAM and disk on what you allocate. Small users can buy prepaid packs from $5 instead.
Together Code Sandbox: Together lists Code Sandbox at $0.0446 per vCPU-hour and $0.0149 per GiB-hour, but the VMs must be enabled for you, so the table prices Code Interpreter: $0.03 per 60-minute session from prepaid credits.
Together Code Sandbox: Plans start at $20 a month, and the whole fee becomes usage credit. You pay per second while a sandbox runs, and a stopped sandbox costs nothing and keeps its disk.
Limits in those scenarios
The Together column is Code Interpreter, and its limits decide every row: sessions last 60 minutes, nothing is kept after a session, it runs only Python and the machine size is not published, so it fits none of the scenarios. Tensorlake Pro fits all five, with up to 1,000 sandboxes at once, but an ephemeral sandbox ends after 10 idle minutes by default and the prepaid plan caps sessions at 24 hours. On bochka the limit that bites is starts: 1,000 short jobs at once need more starts per hour and per day than even the $2,000 plan allows.
| Scenario | Tensorlake | Together Code Sandbox | bochka |
|---|---|---|---|
| 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 want to sign up today and fan out many rollouts from one warm, running state, or need GPUs, outbound rules or sandboxes in your own cloud.
- Warm clones from a running sandbox. For fanning out from one prepared state: Tensorlake copies a running sandbox, memory included, without stopping it first.
- Self-serve, with GPUs and your own cloud. You can sign up and start today, ask for GPU sandboxes, and run in the US, the EU or your own cloud account.
- Outbound rules per sandbox. Useful for untrusted code: each sandbox can allow or deny outbound traffic by domain, IP or CIDR, or have no internet.
Use Together Code Sandbox when
you already run models on Together and want light sandboxes from a tenth of a core, or a simple Python runner billed per session.
- Sandboxes next to your models. Handy if you already run inference or fine-tuning on Together: sandboxes and models share one account and API key.
- Very small machines. Good for many light jobs: a Together sandbox can be as small as a tenth of a core.
- A simple Python runner. For plain Python, the Code Interpreter API runs code in sessions billed per 60 minutes, with no VM to manage.
Where bochka fits
If your agents do real work for hours, with builds, Docker, emulators, a browser and servers other people reach, look at bochka. Each sandbox is a full VM you can start today, with coding agents built in and keys kept outside the machine.
Run emulators and VMs inside
Tensorlake sandboxes set SANDBOX_KVM_AVAILABLE=0 and Together documents no nested virtualization, so an Android emulator or a VM does not start. bochka exposes /dev/kvm on every machine.
Agents and keys built in
On both you install and wire up the agent yourself, and keys you pass in sit inside a machine the agent controls, where one prompt injection can leak them. bochka preinstalls the agents and keeps real keys in a proxy outside.
Reach it like a server
Tensorlake's proxy carries only HTTP, WebSocket and gRPC, and on Together every port is public with preview URLs still coming soon. bochka gives each sandbox token-protected HTTPS on up to 50 ports and raw TCP.
Use bochka when your agents do real work for hours: installs, builds, Docker, emulators, a desktop and servers other people reach, with coding agents built in and keys kept outside the machine. Compare it one to one: bochka vs Tensorlake and bochka vs Together Code Sandbox.
Questions people ask
Is Tensorlake cheaper than Together Code Sandbox?
The only self-serve Together price is Code Interpreter, which fits none of our scenarios because sessions end after 60 minutes and keep no disk. It looks cheaper for small runs, but costs more at scale: $900 against $293 on Tensorlake for sandboxes for your users, and $6,000 against $1,127 for 1,000 short jobs at once.
Which is faster, Tensorlake or Together Code Sandbox?
There is no comparable CPU measurement for either one. bochka scored 26.69 runs per second on the StarSling hpc-sandbox-benchmarks Node.js test, and neither Tensorlake nor Together was part of that run.
Can Tensorlake or Together Code Sandbox run Docker and Android emulators?
Tensorlake runs Docker on its ubuntu-systemd image but has no /dev/kvm, so emulators and VMs do not start. Together's new SDK documents neither Docker inside a sandbox nor nested virtualization. bochka runs Docker, systemd and root and exposes /dev/kvm on every machine.
Can I clone a running sandbox?
On Tensorlake, yes: a running or suspended sandbox can be copied into many warm clones with its memory. Together has no live fork; the sandbox ends with a snapshot and you start copies from it. bochka forks a running sandbox's disk while the original keeps running.
Can I sign up for Together Code Sandbox right away?
Not for the VMs: the together-sandbox SDK and CLI must be enabled for your organization by Together. Code Interpreter is self-serve on prepaid credits. Tensorlake and bochka both let you sign up and start a sandbox yourself.
Which one has GPUs and US regions?
Tensorlake offers GPU sandboxes by request and runs in the US and the EU, or in your own cloud through BYOC. Together does not publish GPU sandboxes or regions for Code Sandbox. bochka has no GPUs and runs only in the EU.
Try bochka
Install the CLI, then run bochka new.
Product preview