Sseekdeploy
Submit a tool

Daytona

Sandboxes for AI agents (containers, VMs, Windows, macOS and GPUs), billed per second for what you reserve.

Paid#sandboxes#ai-agents#vms#gpu#sdk
Visit website ↗Documentation

What is Daytona?

Daytona runs sandboxes for AI agents: isolated computers that code creates, drives and deletes. The default is a container sandbox (1 vCPU, 1 GiB RAM, 3 GiB disk, up to 4 vCPU / 8 GiB / 10 GiB without asking). There are also Linux VM and Windows sandboxes that can pause, fork and snapshot, macOS sandboxes, and ephemeral GPU sandboxes. Agents work through Python, TypeScript, Ruby, Go and Java SDKs, a REST API, a CLI or an MCP server, in shared US and EU regions, dedicated regions, or on your own machines.

Daytona Platforms Inc. bills per second for the vCPU, memory and disk a sandbox reserves, with $200 of free credit. It began as open source, but core development moved to a private codebase in June 2026 and the AGPL-3.0 daytonaio/daytona repo (about 71,000 stars) is archived: forkable, never updated. The SDKs stay Apache-2.0 in the public daytona/clients repo.

On 20 October 2026 Daytona raised a $60M Series B led by FirstMark Growth at a $1B valuation. Its closest rivals here are boat (flat, cheaper VMs, EU only) and Modal Sandboxes (billed on usage).

Our review

3.5

A capable, well-funded sandbox platform with the widest menu in this category, but no longer the open-source project many people first chose it for. One SDK gets you container, Linux VM, Windows, macOS and GPU sandboxes, in US or EU regions or on your own hardware, and in StarSling's October run its VM class came 2nd of 15 on a real-world dependency install. The bill is the catch. Daytona charges for what a sandbox reserves, so a 4 vCPU / 8 GiB sandbox costs $0.33 an hour while it's started, busy or idle: about nine times boat's list price for the same size. New accounts also get restricted internet until a $500 top-up. Since June 2026 the core platform is private and the AGPL repo is archived, so self-hosting today's Daytona is off the table; only the SDKs stay Apache-2.0. Pick it for breadth and maturity, not for price or openness. There's very little independent discussion of it yet.

Pros

  • + The broadest set of sandbox types we list: containers, Linux VMs with pause, fork and snapshots, Windows, macOS, and GPU sandboxes up to 8 GPUs, all from one SDK
  • + Shared US and EU regions, dedicated regions, and bring-your-own-compute regions for data locality
  • + Strong independent numbers for its VM class: 2nd of 15 on StarSling's Mastra cold install (October 2026), statistically tied for 2nd on Node.js CPU throughput

Cons

  • − Billed on reserved vCPU, RAM and disk, not usage: $0.3312 an hour for 4 vCPU / 8 GiB whenever the sandbox is started, about nine times boat's $0.036
  • − Tier gating: Tier 1 caps an organisation at 10 vCPU and restricts outbound internet, and open internet needs Tier 3 (a $500 top-up)
  • − The core platform went closed-source in June 2026; the AGPL-3.0 repo is archived, so the self-hosted path gets no fixes

Reviewed by SeekDeploy Editorial · updated 2026-11-15. Reviews are independent — we don’t accept payment for coverage or ratings.

Should you use Daytona?

A good fit if you're…

  • Teams whose agents need more than one Linux box: Windows desktops, macOS, or a GPU behind the same SDK
  • Products creating many short-lived code-execution sandboxes (code interpreters, evals, RL rollouts) that want ephemeral sandboxes and high creation rate limits
  • Companies that need sandboxes in the US or EU, in a dedicated region, or on their own machines (bring your own compute)

Probably not for you if you're…

  • Anyone who wanted Daytona to self-host it: the open-source repo is archived and the current platform is private
  • Long-running agents on a budget: reserved billing makes an always-on 4 vCPU / 8 GiB sandbox about $242 a month
  • Kicking the tyres on the free credit with open internet: Tier 1 and 2 restrict outbound traffic and you can't override it per sandbox

Pick something else if…

  • you want full Linux VMs for long agent sessions at a flat, much lower hourly price and can live with EU-only regions and no GPU → boat
  • your sandboxes are bursty or your agents' work is Python and GPU-heavy, and you'd rather pay for CPU and memory actually used than reserved → Modal
  • you want an open-source (Apache-2.0) sandbox runtime you can self-host, at the same per-second vCPU and RAM rates, and can do without GPU, Windows and macOS sandboxes → E2B

Gotchas to know before you commit

  • Auto-stop is on by default after 15 minutes without activity, and a process running inside the sandbox doesn't count as activity. A long LLM or build job can stop mid-run; set the auto-stop interval to 0 or refresh activity through the API.
  • Billing follows lifecycle state. Started, creating and stopping bill everything; stopped bills disk only; stopped container sandboxes are archived after 7 days by default and then cost nothing. Snapshots made from a sandbox keep billing for storage after it's deleted.
  • Only VM sandboxes can pause, fork and take hot snapshots, and they can only be created from an existing VM snapshot (no declarative builder). The default container class can't do any of that, and it's the class StarSling's benchmark didn't measure.
  • Tier 1 and 2 organisations get restricted outbound internet that sandbox-level allow lists can't override. Open internet starts at Tier 3, a $500 top-up.
  • Inside a sandbox, nproc and free report the host's CPUs and memory, not your limits. Read /sys/fs/cgroup/cpu.max and memory.max instead, or a build tool may over-parallelise.

Daytona pricing

Checked Nov 15, 2026 against the official pricing page

Pay-as-you-go from a prepaid wallet, per second, for the vCPU ($0.0504/h), memory ($0.0162/GiB/h) and disk a sandbox reserves, with $200 of free credit. It gets expensive when large sandboxes stay started: you pay for the reservation whether the agent is working or idle, so a 4 vCPU / 8 GiB sandbox left on costs about $242 a month. Higher limits and open internet are unlocked by top-ups, not plans.

Pay as you go (Tier 1)
$0 + usage
New organisations with a verified email
  • $200 of free compute, no credit card
  • vCPU $0.0504/h, memory $0.0162/GiB/h, disk $0.000108/GiB/h after the first 5 GiB
  • Per-second billing; stopped sandboxes pay for disk only
  • Limit: 10 vCPU across all running sandboxes; 300 creations a minute
  • Limit: Outbound internet restricted (package registries stay reachable) and can't be overridden per sandbox
  • Limit: Free credit can't be spent on GPU sandboxes
Higher tiers (top-ups)
Usage, same rates
Teams running many sandboxes
  • Tier 2: card + $25 top-up, 100 vCPU, spot GPUs
  • Tier 3: $500 top-up, 250 vCPU, open internet by default
  • Tier 4: $2,000 top-up every 30 days, 500 vCPU
  • Limit: Per-sandbox maximum of 4 vCPU / 8 GiB / 10 GiB unless raised
  • Limit: Windows sandboxes $0.0858 per vCPU-hour
GPU sandboxes
From $0.99 / GPU-hour on demand
Agents writing CUDA, graphics or ML code
  • RTX 4090 $0.99/h, L40S $1.99/h, H100 $3.95/h, B300 $6.25/h on demand
  • Preemptible rates from $0.57/h (RTX 4090) to $4.08/h (B300)
  • Limit: Ephemeral: deleted when stopped
  • Limit: Preemptible (spot) GPUs need Tier 2 or higher
Enterprise / startups
Custom
  • SSO, audit logs, larger limits, bring your own compute
  • Startup programme: up to $50k in credits
  • Limit: Contact sales

What it costs in practice

One agent, workdays
~$58 / month
One 4 vCPU / 8 GiB sandbox started 8 h a day for 22 days = 176 h × (4 × $0.0504 + 8 × $0.0162 = $0.3312) = $58.29. Disk while stopped adds well under $1 (10 GiB, first 5 free). The $200 credit covers about three such months. Left on 24/7 it's 730 × $0.3312 = $241.78.
Code-interpreter calls
~$67 / month
1,000 short sandboxes a day at the default 1 vCPU / 1 GiB, each started for 2 minutes = 1,000 h a month × $0.0666 = $66.60. That's about 0.7 creations a minute on average, far under Tier 1's 300.
Ten agents, workdays
~$583 / month
10 sandboxes at 4 vCPU / 8 GiB, 8 h × 22 days = 1,760 h × $0.3312 = $582.91. Forty vCPU at once is over Tier 1's 10, so this needs Tier 2 (card + $25 top-up), and Tier 3 ($500 top-up) for open internet. The same setup on boat is about $63.

Estimates in USD from list prices. Your bill depends on usage, region, and discounts.

Health & activity

Updated Nov 15, 2026 · last 90 days
Closed sourceThe product itself isn't public. The figures below are for its public companion repo, so treat them as a partial view.
GitHub stars
16
21 forks
Commits
208
at least 12 committers
Merged pull requests
241
Issues opened / closed
53 / 18
median 2.2 days to close
Last push
Nov 13, 2026
Open issues & PRs
44

From daytona/clients (daytona/clients: the public SDKs (Apache-2.0) and CLI (AGPL-3.0). The original daytonaio/daytona repo is archived.). Counts cover the last 90 days unless noted.

Downloads

Benchmarks

Mastra cold install (StarSling sandbox benchmarks)

Source: StarSling hpc-sandbox-benchmarks, run 37073680030

Wall-clock time for a cold dependency install of the Mastra repository on a 4 vCPU / 8 GiB sandbox, median across 12 sandboxes · s, lower is better

  • Namespace37.2 s
  • Daytona (VM)39.3 s
  • Blaxel42 s
  • boat44.4 s
  • Modal (VM)46.4 s
  • Vercel Sandbox59 s
  • Modal (gVisor)70.9 s
  • E2B72.9 s
How this was measured

Third-party run by StarSling, a CI-runner company that isn't one of the providers ranked. It requests 4 vCPU / 8 GiB / 40 GB disk from every provider and reports medians with cluster-bootstrap intervals; the raw dataset and methodology are public. StarSling marks this run partial (800 of 840 cells). This is its headline real-world metric. Daytona's VM class (detected as Firecracker) ranks 2nd of 15, behind Namespace; Daytona's default container class reported no data and isn't ranked. Eight of 15 providers shown; copied from the leaderboard on 2026-11-15.

Versions: Daytona Linux VM sandbox as of 2026-10-03

Published Oct 3, 2026 · source

Node.js web tooling (StarSling sandbox benchmarks)

Source: StarSling hpc-sandbox-benchmarks, run 37073680030

V8 Web Tooling Benchmark throughput on a 4 vCPU / 8 GiB sandbox, median across 5 fresh sandboxes (10 trials) · runs/s, higher is better

  • Brezel30.1 runs/s
  • boat28.1 runs/s
  • Namespace24.9 runs/s
  • Daytona (VM)21.5 runs/s
  • Modal (VM)15.5 runs/s
  • Vercel Sandbox12.7 runs/s
  • Modal (gVisor)12.6 runs/s
  • E2B11.1 runs/s
How this was measured

Same run and methodology as the table above, using the versioned Phoronix Test Suite profile. Daytona (VM) ranks 2nd, statistically tied with boat, Namespace, Microsandbox Cloud and Novita. In the same run Daytona (VM) placed 1st of 15 on STREAM Triad memory bandwidth (176,400 MB/s), a table StarSling flags as 'too few sandboxes' for most providers. Only the VM class was measured. Eight of 15 providers shown; copied from the leaderboard on 2026-11-15.

Versions: Daytona Linux VM sandbox as of 2026-10-03

Published Oct 3, 2026 · source

What the community says

Hacker News & GitHub issues · Aug 17, 2026 – Nov 15, 2026

There wasn’t enough discussion in the last 90 days to summarise fairly. Here’s what there was.

On Hacker News · 5 threads, 14 comments mentioning it

Alternatives to Daytona

boat

Hosting & Deployment
3.5

Persistent Ubuntu VMs for AI agents, billed per second, running in EU data centres.

Free planReviewed

Modal

Hosting & Deployment
4.0

Serverless Python and containers on CPU and GPU: functions, inference, batch jobs and agent sandboxes, billed per second.

Free planReviewed

Fly.io

Hosting & Deployment
4.0

Run full containers close to users, with real VMs and a global Anycast network.

Usage-basedReviewed

E2B

Hosting & Deployment
4.0

Open-source Firecracker sandboxes for AI agents, billed per second while they run, with pause, resume and fork.

Open sourceFree planActiveReviewed

Spotted something wrong or out of date? Flag a correction on GitHub. Health and community data refresh on the 1st and 15th of every month.