Daytona: Sandboxes for AI-Generated Code, Explained
Daytona provides sandboxes where AI agents run the code they write. Each sandbox is a full, isolated computer with its own kernel, filesystem, network and allocated CPU, memory and disk, and Daytona says it goes from code to execution in under 90 milliseconds. Sandboxes can run indefinitely, be snapshotted mid-task, forked into parallel branches and resumed later, and Daytona also offers Linux, Windows and macOS desktops for agents that need to use a computer like a person. Its customers include LangChain, Writer, Turing and SambaNova, and it raised a $24 million Series A in February 2026.
The quick answer: Daytona is one of the main sandbox platforms for AI agents, alongside E2B, with strengths in long-running, stateful work and the ability to fork and snapshot environments. It suits teams building coding agents, evaluations and reinforcement learning setups. One thing to know: Daytona’s GitHub repository, with about 71,000 stars, stopped being updated in June 2026 when development moved to a private codebase. Everything below comes from Daytona’s site, pricing, documentation, GitHub repository and Series A announcement.
Daytona at a glance
Checked against Daytona’s site, pricing page, developer summary, GitHub repository and Series A post on November 13, 2026.
| Daytona | |
|---|---|
| What it is | Isolated, stateful sandboxes for running AI-generated code and agent workflows |
| Startup time | Under 90 milliseconds from code to execution, per Daytona |
| Capabilities | File, Git, language server and execution APIs; snapshots, forking, SSH, browser terminal |
| Desktops | Linux, Windows and macOS, controllable by code |
| Access | SDKs, REST API and CLI; Docker-compatible images |
| Pricing | Usage-based, billed per second; $200 in free compute |
| Company | Co-founded by Ivan Burazin (CEO); Series A led by FirstMark Capital |
What Daytona does
Daytona’s developer summary calls it “a secure and elastic infrastructure for running AI-generated code” that provides “isolated sandboxes with dedicated compute, filesystem, and network resources for AI agent workflows.”
Its Series A announcement explains the idea behind it. Today’s cloud was built for production software, “stateless, immutable, optimized to run the same code the same way every time.” Agents, like people, need computers where they can “install packages, run experiments, make mistakes, and recover,” but at a scale and speed people never needed. Daytona thinks of sandboxes as “programmatic, composable computers” whose CPU, memory, storage, GPUs, network and operating system can be combined on demand, then started, paused, forked, snapshotted or destroyed at any point.
In practice, a sandbox gives an agent:
- Code and command execution in Python, TypeScript and JavaScript, with real-time output.
- File, Git and language-server APIs, so a coding agent can edit, commit and get code completion.
- Persistence, with sandboxes that “Run Indefinitely” for long tasks, and snapshots to save and restore any workflow.
- Forking, so an agent can split into parallel branches to try different approaches, keep the good ones and discard the rest.
- Human access, through SSH, a browser terminal and one-click opening in a code editor, for debugging or stepping in.
- Shared data across sandboxes without breaking isolation.
Desktops for computer use
Beyond code, Daytona offers “Secure virtual desktops (Linux, Windows, macOS) you can control with code.” The Linux desktop comes with full root access, the macOS desktop is aimed at iOS development and mobile app automation, and the Windows desktop handles Windows-specific automation and testing. Agents can operate them by seeing the screen and using a mouse and keyboard, while people can watch or intervene.
Who uses it
Daytona’s Series A post says it “runs agent workloads for early-stage YC companies and Fortune 100 enterprises alike,” naming LangChain, Turing, Writer and SambaNova. Its main uses are running AI-generated code safely, running evaluations across many parallel environments with reproducible snapshots, and reinforcement learning, where agents need long-running, stateful environments to learn in. It also supports data analysis that renders charts and plots.
The company says it reached a $1 million forward revenue run rate in under three months and doubled it six weeks later.
The open-source change
Daytona built much of its developer following as an open-source project; its GitHub repository has about 71,000 stars. In June 2026, the repository’s README was updated to say: “This repository is no longer maintained.” Daytona’s core development moved to a private codebase, and the public code will get no further updates, fixes or releases, though it stays available to use and fork under its license as it was at version 0.190.0.
For teams choosing Daytona today, that means it’s effectively a managed service. If self-hosting on maintained open-source code matters to you, that’s a meaningful change from its earlier positioning.
Pricing
Daytona’s pricing is usage-based and billed per second, with $200 in free compute to start and no credit card required for the trial:
| Resource | Price |
|---|---|
| vCPU | $0.0504 an hour |
| Memory | $0.0162 per GiB an hour |
| Storage | $0.000108 per GiB an hour, after the first 5 GiB free |
| Windows | $0.0858 per vCPU an hour |
| GPUs | From $0.57 an hour (preemptible RTX 4090) to $6.25 an hour (on-demand B200 or B300) |
A sandbox with 1 vCPU and 1 GiB of memory works out to roughly $0.067 an hour. Volume discounts are available, and Daytona runs a startup program.
The company
Daytona’s CEO and co-founder, Ivan Burazin, previously built the cloud development tool Codeanywhere, founded the Shift developer conference, and served as chief developer experience officer at Infobip. The company started out managing development environments for programmers before refocusing on infrastructure for AI agents.
In February 2026, Daytona raised a $24 million Series A led by FirstMark Capital, with FirstMark’s Matt Turck joining its board. Angel investors in the round include Gorkem Yurtseven of Fal, Theo Browne of T3 Chat, Eno Reyes of Factory and Nikita Shamgunov of Neon, according to Tech.eu.
How Daytona compares
Daytona’s closest competitor is E2B, which also runs isolated sandboxes for AI agents at large scale and keeps its runtime open source under Apache 2.0. Daytona emphasizes long-running persistence, forking and snapshots, GPUs and Windows and macOS desktops. Orgo focuses on persistent desktops for computer-use agents, Cua on running agents on computers you own, and Tensorlake on sandboxes paired with versioned storage and hosted Git for agents. For agents that only need a browser, Browserbase is simpler.
Sandboxes matter when agents write and run their own code. For everyday work in the tools you already use, Carly connects to thousands of apps directly, so from ChatGPT, Claude or a text message it can sort email, book meetings and keep your CRM current.
Who Daytona suits
Daytona is a good fit if:
- Your agents write and run code and need isolated, persistent environments.
- You want to fork and snapshot agent work, for evaluations or reinforcement learning.
- You need GPUs, or Windows and macOS desktops, alongside Linux sandboxes.
- You want fast startup and per-second billing.
Look elsewhere if:
- You need actively maintained open-source code to self-host.
- Your agent only needs a browser.
- You want a finished assistant rather than infrastructure.
FAQ
What is Daytona?
Daytona is infrastructure for running AI-generated code. It provides isolated sandboxes, each a full computer with its own kernel, filesystem and network, that start in under 90 milliseconds and can persist, snapshot and fork.
Is Daytona open source?
Daytona’s GitHub repository remains public under its license, but as of June 2026 it’s no longer maintained. Core development moved to a private codebase.
How much does Daytona cost?
Daytona is billed per second by usage: $0.0504 per vCPU hour, $0.0162 per GiB of memory per hour, plus storage and optional GPUs. New accounts get $200 in free compute.
Does Daytona support computer use?
Yes. Daytona offers Linux, Windows and macOS desktops that agents can control with code, seeing the screen and using a mouse and keyboard.
Who founded Daytona?
Ivan Burazin is Daytona’s CEO and co-founder. The company raised a $24 million Series A led by FirstMark Capital in February 2026.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
