Introducing Together Code Sandbox & Together Code Interpreter: SOTA code execution for AI

LLMs have revolutionized the way we generate code, but running that code in a secure, customizable, and performant environment remains a challenge. Developers face delays as they manually set up environments, debug generated code, and configure complex infrastructure.

Today, we're excited to announce two major products designed to bridge this gap:

• Together Code Sandbox: Customizable VM sandboxes to build full-scale development environments for AI.
• Together Code Interpreter: Session-based code execution in isolated Sandboxes via API calls that stream results.

While both products run on the same robust infrastructure, Together Code Interpreter was purpose-built for use cases that require straightforward code execution without the advanced customization of Together Code Sandbox.

With this launch, we now offer a suite of dev tools to transform LLM generated code into the solutions your customers need—all powered by secure, mature infrastructure that is ready to scale to millions of users in production.

Together Code Sandbox: fully managed dev environments in seconds

Together Code Sandbox provides customizable, instantly bootable microVM environments powered by robust, mature infrastructure.

With industry-leading performance and infrastructure proven in demanding large-scale deployments, it enables developers to build full-scale development environments for AI and scale them effortlessly to millions of users.

• Fast and scalable snapshotting: Tried-and-true memory snapshotting technology that starts VMs from a snapshot in 500 ms (P95) and clones them in under 1 second.
• Suite of dev tools: Control sandbox dev environments with tools for terminal access, tasks, preview host, sessions, and more.
• VM scaling: Hot-swappable VM sizes ranging from 2 to 64 vCPUs and 1 to 128 GB RAM.
• Persistent storage: A git-versioned filesystem ensures your data remains accessible and secure.

To make it easier to navigate all these tools, we created a simple playground you can use to see the capabilities of Together Code Sandbox in action. Give it a try or catch a sneak peek below.

AI pioneers have been leveraging Together Code Sandbox to accelerate their go-to-market and drastically improve code environment performance. Thanks to this flexible infrastructure, HeroUI cut the development time of HeroUI Chat from 5 months to 2 weeks, reducing their VM startup times from 2 minutes to under 2 seconds.

“With Together Code Sandbox, we had an MVP running in a couple of hours, so it finally allowed us to test the result of HeroUI Chat, instead of waiting five months to see the result in a preview environment. Now, this is the core of HeroUI Chat; it is the platform that allows us to run the projects. Without it, we don't have a way to show you the result of the AI.” — Junior Garcia, Founder and CEO of HeroUI

To empower developers working on projects of any scale, we are launching Together Code Sandbox with transparent, competitive pay-as-you-go pricing:

• $0.0446 per vCPU per hour, plus
• $0.0149 per GiB RAM per hour

Start building with Together Code Sandbox by following our Docs and don’t miss our deep-dive blog post for additional details on its features and use cases.

If you’re interested in a custom large-scale deployment, contact our Sales team for information on custom pricing.

Together Code Interpreter: run code with a simple API call

Whenever you need simple code execution, Together Code Interpreter (TCI) offers an easy-to-use API for running Python code.

With this API, you can pass in LLM-generated Python code and return the executed code. Optionally, you can reuse an active session that will hold state for the duration of its 60-minute lifespan. 

• Instant code execution: Run Python scripts with a single API call (support for more languages coming soon).
• Session-based: Each execution session lasts 60 minutes and can be called multiple times.
• Secure and isolated: Robust sandbox environment protects sensitive data and ensures secure execution.
• Simple pricing: Each 60-minute session is priced at $0.03.

{{custom-cta-1}}

One particularly valuable application of TCI is in reinforcement learning. During training, TCI can be implemented to run unit tests on model-generated code, which is then used to compute rewards for code RL training.

Agentica, an open-source initiative from Berkeley AI Research and Sky Computing Lab, integrated TCI into reinforcement learning operations while training DeepCoder-14B-Preview, running 1024 code executions in parallel to significantly accelerate training cycles.

“Together Code Interpreter has dramatically accelerated our RL post-training cycles, enabling us to reliably scale to over 100 concurrent coding sandboxes and run thousands of code evaluations per minute. Its reliable and scalable infrastructure has proven invaluable.” — Michael Luo & Sijun Tan, Project lead at Agentica

You can start using Together Code Interpreter today by using our Python SDK or our API. For more details about TCI, check out our deep-dive blog post.

Together Code Sandbox vs. Together Code Interpreter: choosing the right fit

In summary, while Together Code Sandbox and Together Code Interpreter are powered by the same robust infrastructure, they cater to different use cases and needs.

Below is a short comparison table to guide you in selecting the right fit for you.

To further elaborate on the most common use cases, here is a breakdown of which solution is ideal for different developer workflows:

• Agentic apps: Use TCI for instant execution of Python code generated by LLMs.
• RL training pipelines: Integrate TCI for batch testing.
• AI IDEs or SaaS platforms: Deploy fully managed development environments using Together Code Sandbox, complete with interactive shells and real-time previews.

Get started

This launch marks a big milestone in solving some of the toughest challenges of LLMs when it comes to code: isolation, security, performance and scalability. With Together Code Sandbox and Together Code Interpreter, we are further accelerating how developers train, build, deploy and scale AI coding products.

We have prepared quickstart guides to help you get started with Together Code Sandbox and Together Code Interpreter, and don’t hesitate to contact our Sales team if you’re interested in learning more about how you can leverage these products at scale.