We added support for interactive charts. E2B now automatically detects charts when executing Python code with runCode() in JavaScript or run_code() in Python. The Python code must include Matplotlib charts.

E2B sends the data of the chart back to the client, so you can then use your favorite chart library and create custom interactive charts.

See corresponding docs page, or try interactive charts in the E2B Analyst app.

E2B Artifacts

• Add support for Gemini Pro 1.5 002, Gemini Flash 002, and Mistral Small models

Infrastructure

• Add proper sandbox duration limit to the API

Core & Code Interpreter SDK

• Fix RPC issue causing Server disconnected without sending a response

This week we improved the E2B Artifacts and built examples that test the capabilities of OpenAI's o1 model.

E2B Artifacts

• Lift o1 rate-limits

• Allow to bring your own API keys

• Fix issue with E2B API keys not set correctly for recurring users

• Add rate-limit adjustable via env variables

Cookbook

We built an example that lets the o1 model write the code to train a machine learning model on a dataset from Kaggle.

We used the E2B Code Interpreter SDK for running the LLM-generated code tasks in a secure and isolated cloud environment. As LLMs, we used o1-mini to generate a thorough plan for the task with many code blocks, and GPT-4o-mini to extract the final code from the plan generated by o1-mini.

The example has a Python and JS/TS version.

Core & Code Interpreter SDK

• Indicate data from multiple plots in the response

• For histograms, indicate which axes the bar data should be displayed on

• Return tick labels in better format

• Add graph types + improve output data structure

• Propagate kernel restarts to user

• Add units in other contexts (e.g. tool tip)

• Fix terminal pty missing output

• Add terminal (pty) support to Python SDK

E2B Artifacts

• Update artifacts support for o1-preview and o1-mini

• Fix password reset

Infrastructure

• Fix deleting templates without successful build

Core & Code Interpreter SDK

• Ensure sequential execution when executing code in Code Interpreter​

• Separate Python environment for the proxy webserver for Code Interpreter​

• Extract data from notebook to beta SDK​ to allow user easily accessing data from DataFrames

• Fix an issue causing new kernel to not respect the provided working directory

E2B Artifacts

• You can now iterate on the generated artifact by chatting with the LLM

• You can upload image to the models that support multi-modal input

• Added button to start new chat

• You can now download generated artifacts

• Improve startup speed of the Next.js template

E2B Artifacts

• Improved UX

• Add your own personas and LLMs/providers

• Out of the box support for shadcn components

Infrastructure

• Fixed a critical memory issue causing sporadic 504 and 503 errors for users

• Moved UFFD handling to orchestrator

• Fixed a bug with starting process with PTY more than once

• Increased limit for number of open files and connections

E2B Artifacts

• Added support for Vue.js, Gradio

• Added tool and output streaming

• General UI & UX improvements

• Improve the general reliability of the generated apps

Infrastructure

• Fixed a memory bug causing 504 and 502 errors when interacting with the sandbox

This week we worked on the core SDK and made improvements in the E2B Artifacts.

• Dashboard #421: Broken CSS rules on Safari

• Beta Core SDK #423: Add download_url/downloadUrl method to sandbox

• Beta Core SDK #425: Env vars

• Infra #177: Add possibility to setup env vars on sandbox.create()

• We also added stderr to process exception in Beta Python SDK.

  
  

• Artifacts #14: Add integration for Groq API

• Artifacts #17: Add license

• Artifacts #20: Don't wait for Nextjs page to compile

• Artifacts #21: Fix Supabase on localhost

• Artifacts #24: Added model support and UI picker

• Artifacts #25: Make Supabase opt-in instead of opt-out

We launched open-source E2B Artifacts. You can use E2B artifacts to build various types of apps, just by describing what you want. Thanks to running the code in the E2B sandbox, the app is built right in front of you.

Examples of artifacts that you can generate are Next.js app, data visualization charts, Streamlit apps, or interactive tables.

E2B Artifacts are fully open source.

New cookbook examples

• Add code interpreter examples for Together AI

Fixes and improvements

• CLI #417: Allow user to set a default team and run the list, kill, delete commands with the context of different user's teams

• Docs #418: Add sections about the —-team flag to relevant commands

• Infra #172: Fix a bug that prevented users to set the sandbox's timeout duration lower than the default one

We worked on new features in the Code Interpreter SDK, and on improving the infrastructure.

We released a guide for migrating from our current SDK to the new beta SDK.

Additionally, we continued working on scaling and towards the production release of the new SDK.

Fixed bugs and improvements

• Infra #166: Set up proxy timeouts and improve DNS resolving

• Infra #130: Add health check alert for Nomad and Consul

• SDK #364: Fix bug with permissions when creating a new directory

• SDK #375: Make CodeInterpreter and Sandbox constructors private

We released the new E2B dashboard. It allows users to easily monitor their usage & costs, and manage team members. Go ahead and try it at e2b.dev/dashboard.

Fixed bugs and improvements

• Docs #407: Update PostHog package and fix the api_host

• Docs #406: Use Buffer.from() when writing a file

• Docs #402: Merge reverse proxy and docs into one app

• Code Interpreter SDK #30: Improve Jupyter proxy server implementation

This week we continued working on the new dashboard and SDK.

This week we were working on the new dashboard and SDK.

Fixed bugs and improvements

• Code Interpreter SDK: Fixing example paths

This week we made an open source version of Anthropic's Artifacts, using Claude-3.5 Sonnet, Next.js, and E2B for the AI code execution. See the code here.

Web-scraping assistant with Firecrawl

Another new open-source example is an agent that scrapes a website and plots the most common topics. It was added by Eric Ciarla, and uses Firecrawl for web scraping, Claude-3.5 Sonnet by Anthropic, and E2B Code Interpreter SDK.

Try the example here.

Fixed bugs and improvements

• Infra: Network speed improvements

This week we built a simple AI agent with Firefunction-v2, the new LLM by Fireworks AI. We gave it code interpreting capabilities with the E2B Code Interpreter SDK and used it for chart generation.

Check out the complete E2B Cookbook.

Fixed bugs and improvements

• Infra: Optimize db access

See the public roadmap with planned features here.

We are adding more language runtimes to the Code Interpreter SDK. Now we can handle AI-generated code written in JavaScript, R, and Java, besides the default Python.

You can try this experimental feature here: https://github.com/e2b-dev/code-interpreter/pull/20. Follow the instructions on the link.

Fixed bugs and improvements

• E2B SDK: CLI add build args

See the public roadmap with planned features here.

We added a JavaScript & TypeScript version of AI data analyst with Mistral's Codestral. It plots a chart based on a CSV file, using the E2B Code Interpreter SDK as the code execution layer.

Fixed bugs and improvements

• E2B SDK: Add max retries to websocket reconnect in Python SDK

• Code Interpreter SDK: Move start cmd to envd

• Infra: Refactor new user triggers

• Infra: Manage start command in envd

See the public roadmap with planned features here.

We created a new example with AI data analyst that uses Codestral and has full code interpreter capabilities given by the E2B Code Interpreter SDK. The analyst is working with a CSV file and generating a chart based on the data.

See the complete code in our Cookbook and if you have an idea for your own project with E2B, please open an issue.

Fixed bugs and improvements

• Infra: Don't log internal logs in logs collector

• Infra: Allow specifying sandbox timeout when creating the sandbox and modifying it later

See the public roadmap with planned features here.

This week we worked on improvements to our infrastructure, for example:

• Increase limit for file upload

• Fixed the bug related to closing sandboxes.

See the public roadmap with planned features here.

GPT-4o with code interpreter

We created two examples with GPT-4o, one in JS/TS and one in Python. We gave the newest OpenAI model full code interpreter capabilities, using the E2B Code Interpreter SDK.

See the complete code in our Cookbook and if you have an idea for your own project with E2B, please open an issue.

Code interpreter analyzing uploaded dataset

We also added a Python example that shows how to upload a dataset and analyze it with the E2B code interpreter and Llama3.

Fixed bugs and improvements

• Code Interpreter SDK: RPC error on 0.0.5, unable to rollback to 0.0.3 or 0.0.4

• Code Interpreter SDK: Fixed problem with running multiple code interpreters in Python

• Infra: Client instance unexpectedly restarts

See the public roadmap with planned features here.

New example: Anthropic code interpreter in JS/TS

We gave Claude a full code interpreter capabilities with just a few lines of code, using our Code Interpreter SDK. Features:

• Streaming

• Internet access

• Python & JS SDK.

See the complete code in our Cookbook and if you have an idea for your own project with E2B, please open an issue.

Fixed bugs and improvements

• E2B SDK: Enforce calling create method in JS SDK

• Infra: Add start command handling to the API and env-build-driver

See the public roadmap with planned features here.

We released the Code Interpreter SDK. It was made as a building block for AI engineers:

• Works with any LLM and AI framework

• Supports streaming content like charts and stdout, stderr

• Python & JS SDK.

The SDK is 100% open source.

See our Cookbook for the examples of what to build with the Code Interpreter SDK.

Fixed bugs and improvements

• Code Interpreter SDK: Improve exports

• Code Interpreter SDK: Allow serialization of Execution object

• Infra: Improve envd loggers so sandbox logs can be easily filtered

• Infra: Create public status page

• E2B SDK: Add CLI command for printing logs from sandbox

• E2B SDK: Remove runCode and runCmd from SDKs

  
  

See the public roadmap with planned features here.

We added "Hello World" guides to our documentation showing how to add code interpreting to LLMs using our Code Interpreter SDK.

• Hello world in TypeScript

• Hello world in Python

You can find the complete list of examples in our Cookbook.

Fixed bugs and improvements

• SDK: Update metadata.py to not use deprecated Sandbox.create() method

• Docs: Add documentation for code interpreter SDK

• Infra: Add logs and metrics from proxies

• Infra: Add swap and adjust OOM handling in sandbox

• Infra: Fix UFFD and Firecracker killing in orchestrator

• Infra: Don't track sandbox id for monitoring in not prod environment

• Infra: Create public status page

  
  

See the public roadmap with planned features here.

We integrated E2B to OpenDevin. This open-source AI software engineer now supports the E2B sandboxes as an alternative to a Docker container for code execution.

Fixed bugs and improvements

• SDK: Add domain parameter to SDK

• SDK: Add endpoint for status page

• Docs: Add missing parentheses in sandbox.close() call

• Infra: Set limits for orchestrator based on the machine

• Infra: Enable monitoring on instance groups + collect logs for Consul and Nomad

• Infra: Decrease max unavailable servers to 1

• Infra: Add default consul agent token

• Infra: Change nomad connection and scraping limits

• Infra: Remove unnecessary cleanup method and fix cleanup order

• Infra: Update and simplify Nomad jobs

• Infra: Set file descriptors' limits

• Infra: Refactor custom driver for templates into a system job

• Infra: gRPC logging

• Infra: Improve huge pages handling in client start script

See the public roadmap with planned features here.

Fixed bugs and improvements

• New: Dedicated SDK for Code Interpreter

• Infra: Add API endpoint for retrieving sandbox logs

• Bug fix: Fix missing MMDS in some sandboxes

• Bug fix: Fix process output scan buffer initialization and resizing

• Bug fix: process.wait sometimes times out

• Code Interpreter SDK: Move dev dependencies to devDependencies

Product roadmap

See the public roadmap with planned features here.

New example: Anthropic tool calling with code interpreter 💻

We gave the new Anthropic tool calling support a full code interpreter capabilities. We created a Python notebook showing how to use E2B to give Claude model a code interpreter with just a few lines of code. Features:

• Code runs in Jupyter notebook

• Streaming

• Customize libraries

• Internet access

• Python & JS SDK.

See the complete code in our Cookbook and if you have an idea for your own project with E2B, please e-mail us at hello@e2b.dev.

Other fixed bugs and improvements

• New: API server communicates with Nomad via streaming (better performance)

• Bug fix: Remove an old alias if the template already has some

• CLI Bug fix: Make sandbox kill command work with the long form of sandbox ID

• Infra: Fix Consul bug preventing bootstrapping

• Infra: Modify networking setup to stop exposing Nomad metrics and delete whitelisted IP address

Product roadmap

See the public roadmap with planned features here.

Build of docker image for a new template is now done on user's local machine instead of E2B's cloud. Here's how this new approach improves the DX:

• Allows referencing private docker images

• Manages build cache

• Higher iteration speed

• Prevents uploading node_modules or venv by accident

• Option to use a 3rd party service like GitHub Actions to build the sandbox template.

To build a new template users have to upgrade their CLI to the newest version:

npm install -g @e2b/cli@latest

Find more about building your custom templates here.

Other fixed bugs and improvements

• New: Use local machine's Docker instance when building a custom sandbox template image

• Bug fix: Remove extra whitespace when logging to Docker on Windows

• Bug fix: Make sandbox kill command work with the long form of sandbox ID

• Bug fix: Fix clock drift on startup

• Bug fix: Add quotes around access token in CLI

• Infra: Use Consul token for scraping metrics

• Infra: Add placeholder stats channel for tasks

• Infra: Remove a template alias when sandbox template is deleted

• Infra: Deny access to Consul by default

• Infra: Don't cleanup nomad logs on restart

• CLI: Improve cross platform compatibility of e2b template build command

Product roadmap

See the public roadmap with planned features here.

Fixed bugs and improvements

• Fix enforcing the limit of concurrent sandboxes per team

• Improve clock sync handling in FC

• Cleanup kernel building

• Switch the template build to local

Product roadmap

We released a public roadmap on GitHub to share features and bugs that we are working on. If there is anything you would like to see us add, please open an issue on our GitHub.

Multiple consul and nomad servers

We added an option to manage multiple Consul and Nomad servers simultaneously.

Other fixed bugs and improvements

• Export all errors and exceptions

• Raise proper exception on process.wait timeout

• Add temporary limit for concurrent new sandbox requests

• Make client startup script rerunnable

This week we were working on a new CodeInterpreter API. It's still in its experimental phase, but we love to hear your feedback. We added support for multiple Nomad and Consul servers.

Other fixed bugs and improvements

• Change auth buttons to links

• Update CLI commands in the guide to reflect the latest CLI version

• Standardize text format and grammar across CLI commands, args, and options

• Add email authentication

Up to 5x faster sandboxes

We implemented huge pages - a feature that allows customers to use 2 MiB chunks of memory instead of 4 KiB chunks, which are the default.

Huge pages are now enabled by default in all newly created custom sandboxes. If a user has an existing custom sandbox, they must rebuild it by calling e2b build in the directory with the e2b.toml file.

Custom CPU And RAM

We allowed Pro users to customize the number of vCPUs and RAM size for the sandboxes. The sandbox compute now also has a dedicated page in our documentation.

Pro users can now choose:

• 1 - 8 vCPUs

• 128 MiB - 8 GiB RAM

Other fixed bugs and improvements

• Fix upload path for FC versions

• Add command for killing sandboxes in our experimental SDK

• Release CLI v0.3

• Document process for customizing sandbox compute

• Fix type issue in Python 3.8

Getting started with sandboxes

If you want to try the sandboxes, start here and let us know at hello@e2b.dev if you need any support.

We enabled billing on E2B. Read more here about what we charge for and how to monitor your usage. We added the Your Usage page to our Docs to monitor your spending.

Fixed bugs and improvements

• Go context used for tracing is not closed properly

• Change max length based on the team tier

You can now connect your storage buckets to the sandbox. Connecting your cloud buckets is useful for filesystem persistence or giving the sandbox access to your data. This is achieved via the FUSE protocol which is supported by all popular cloud buckets: Amazon S3, Google Cloud Storage, and Cloudfare R2.

Get started here.

Fixed bugs and improvements

• Add FUSE to docs

• reconnect() now accepts API key

• Enable billing for users

• Experimental stateless SDK

• Fix incorrectly deleting template files when deleting templates

• Add API endpoint for instantly killing sandboxes

• Fix the issue preventing using any Debian images because of incorrectly called bash

This week a contributor Keegan McCallum from our community built an example of a code interpreter with AutoGen and E2B sandboxes used for the LLM code execution.

Fixed issues:

• ValidationError for SubscriptionArgs handler

Merged PRs:

• Fix sandbox ID returned from SDKs list methods

• Fix broken provisioning for some Debian versions

Merged PRs:

• List running sandboxes

Fixed issues:

• JS SDK - Reject promises the correct way

• Python SDK Websocket reconnect sometimes takes unnecessarily long

• Unable to create custom templates with Dockerfiles containing more than ~4.3GB of local data

• Kill WebSocket and refresh threads automatically after close or program exit

• Using sandbox.download_file on an empty file throws error

• Building a custom sandbox fails when e2b.Dockerfile is in .dockerignore

• Improve SDK error message when API key is missing

Merged PRs:

• Always include dockerfile when building custom sandbox

• Improve API key is missing error message

• Add check for maximum body upload size when building custom sandbox

• Improve SDK performance - Not waiting for unsubscribe where not needed, subscriptions done in parallel now

• Move refreshing and message processing to daemon threads in Python SDK

This week, we've been working on improving the SDK performance.

We completed the necessary steps towards open-sourcing our infrastructure.

One of our contributors finished the JS/TS version of the AI GitHub Developer.

We've been working on a major update of our SDKs, and improving our infrastructure.

We also fixed the following issues reported by users:

• https://github.com/e2b-dev/E2B/issues/282

• https://github.com/e2b-dev/E2B/issues/279

Product

We were implementing the billing system, and improving the performance of our sandboxes.

New bounties

New bounties (contributions for a financial reward) were opened this week:

• Build a code interpreter with Autogen and E2B

• Build a code interpreter with LlamaIndex and E2B

• Build a code interpreter with Hugging Face and E2B

See the complete info on the bounty program here. If you to discuss anything, contact us in the Bounties channel on our Discord.

Blog post about AI agents

In the newest blog post, we interviewed AI founders and developers. They shared their predictions for AI agents in 2024, their favorite AI product of 2023, and a new year's AI resolution.

This week, we opened a new bounty for our developer community.

The bounty is offered for a JavaScript/TypeScript version of our existing Python AI GitHub developer. The developer serves as a simple demonstration of using E2B sandboxes.

To apply for this bounty or other bounties, write us in the Bounties channel on our Discord.

This further worked on our SDK and built a cookbook example of an AI developer to demonstrate the power of E2B sandboxes “in terrain”.

Improvements

We fixed bugs and improved the following:

• Accelerated the sandbox spawning time

• Fixed Python code output buffering.

• Moved start command handling in Python SDK to a thread.

AI GitHub Developer demo

We released an open-source AI assistant that can create pull requests on any repository. It is built with OpenAI’s Assistants API and E2B sandboxes.
The developer Can clone the repo, edit, read & write files, and is controllable from the terminal.

See the complete code and demo video. The AI developer is not meant as a full product, but rather as an example that can inspire other developers.
We published a step-by-step guide, so check it out if you want to build something similar.

We dedicated this week to further improvements in the E2B SDK and infrastructure.

This week we continued working on the E2B SDK and infrastructure.

This week we worked on fixing bugs in the E2B SDK and improving reliability.

We changed the default disk space to 1 GB for free users.

Support 

Please reach out to us if you need help with E2B.

• Need support or want to share feedback? Join our Discord.

• You can also reach us at hello@e2b.dev.

If you are on X (Twitter), we would appreciate if you support our work there.

This week we made improvements in our SDK, added new features for sandboxes, and fixed bugs.

Start command

Custom sandboxes now allow you to specify a command that will be running when a sandbox is spawned. This way you can run a command that will start your app or any specific process.

When building a sandbox template, you can specify the start command using the -c option.

e2b build -c "npm start"

Delete sandbox template

We released an option for deleting sandbox templates.

e2b delete

This week we are announcing native support for the new OpenAI's Assistants API. That means you can now define actions for the AI assistant and we will automatically execute them inside the E2B Sandbox.

Combined with the custom sandboxes feature, this allows you to create powerful AI assistants with tools completely predefined by you.

To try the OpenAI Assistants with E2B, start here. You can follow our guide in Python or JavaScript.

See the complete code in the E2B Cookbook.

New feature: Reconnect to running sandbox

This week we launched a new experimental SDK feature. If you want to keep the sandbox alive after disconnecting from it, you can now explicitly say how long you want to keep it alive. You can then reconnect to the sandbox later.

Read more here.

This feature is available in versions:

• JS: 0.10.0

• Python: 0.11.0

For implementation assistance or feedback, contact us at hello@e2b.dev or on our Discord.

Docs and SDK improvements

We also made a few updates to our SDK and documentation.

Docs

• Shorter time for code snippets to be ready for running (using custom sandbox)

JS SDK

• Pass sandbox id directly

• Pass command directly to process

Both SDKs

• Add StartAndWait for process.

This week we released the Custom Sandboxes for AI-powered apps and agents.

The E2B Sandbox is an isolated long-running cloud environment where you can let any LLM use tools exactly like you would do locally.

You can now create a Custom Sandbox for any use case, e.g. your own AI coding copilot, data analyst, or Al Browser. The Custom Sandbox allows you to spawn a sandbox with an environment predefined by you and then control it with the E2B SDK.

We also offer two premade sandboxes (Default Sandbox and Code Interpreter Sandbox).

Example of running Python code with the E2B Sandbox

How to start with Custom Sandboxes:

1. Sign in with your GitHub account to get your free E2B API key.

2. Install the E2B SDK, either in JS or Python.

3. Create your custom sandbox (you can follow our guide) or try out one of our premade sandboxes.

   
   

Support 

• Need help or want to share feedback? Join our Discord.

• You can also reach us at hello@e2b.dev.

If you are on X (Twitter), we would appreciate if you support our work there.

We’re releasing a private beta of E2B Custom Sandbox. 

If you want to test them, please contact vasek@e2b.dev or the @e2b-team on our Discord.

Custom Sandboxes in private beta

LLM Sandbox is an isolated cloud environment for LLMs with support for long-running sessions.

You can customize sandboxes by creating a Sandbox Template, which is just a simple Dockerfile. This way, you can preinstall any needed libraries and packages, or prepare any files you need. Then you can simply control your Custom Sandbox with the E2B SDK.

This week, we announced the official E2B integration into LangChain and created a guide for trying out the E2B Data Analysis Sandbox with LangChain.

LangChain integration

E2B is now available as a tool on LangChain. The E2B LLM Sandbox is ideal for building your own AI Data Analysts and Code Interpreters.

A big advantage of the E2B Sandbox is a safe code execution - allowing for example installing packages dynamically, generating charts, or file upload & download.

Bug fixes

• Fix SDK browser compatibility

• Fix installing system packages inside Sandbox

Guides and examples

We wrote a tutorial for using E2B with LangChain to build your own AI data analyst.

We put together more examples of using E2B, which are now shared within the E2B Discord community. 

• Need help or want to share feedback? Join our Discord.

• If you like the guide, please support us with a star on GitHub.

• Follow us on X (Twitter).

• You can also reach us at hello@e2b.dev.

🎃 Happy Halloween hacking!

This week, we fixed various bugs in our SDK and worked on our infrastructure.

New design and communication

We made our docs a bit nicer and better structured. See how to get started with E2B, a useful step-by-step example,  or an overview of the sandbox API. Check out our updated landing page. How do you like the new E2B logo?

Read about AI agents’ SDKs and frameworks

We shared an update on key topics of developing LLM agents and apps, covering for example security, privacy, memory management, frameworks, and developers' perspectives.

You can read it also on Medium or go through the frameworks and SDKs here.

This week, we continued improving the E2B sandboxed environments and the Superagent integration of E2B. To try Superagent with E2B, create a Code interpreter API and then select it for your agent to use.

Additionally, we worked on the backend for building your custom Advanced Data Analysis/Code Interpreter, so stay tuned.

We published more content and expanded our blog to Medium. You can read the newest AutoGen tutorial for code-executing agents.

If you haven’t done it, try out our SDK here and tell us how you liked it via Discord or by tagging @e2b_dev on Twitter. 

• E2B is now integrated into Superagent, meaning there is added support for a Code Execution feature. You can try it here.

• This integration allows the AI agents to actually run code in a sandboxed environment, which helps for example with creating charts, doing requests, or analyzing files.

• Join the Superagent Discord to stay updated and discuss your creations.

The code execution API in Superagent is still in its alpha version, as we are still working on improvements. We are very grateful for your feedback - let us know on Discord or contact our CEO at vasek@e2b.dev if you have any questions.

More E2B updates

• We updated the E2B SDK - you can now set it on stdout, on stderr, and on exit for the whole session.

This week, we released a simple way to run code generated by LLMs in the cloud:

1. Import e2b

2. Select runtime and pass LLM-generated code

3. Run code in a sandboxed environment

4. Get stdout and stderr

This way, developers of AI agents and apps don’t need to handle any infrastructure.

Read more in our documentation and let us know on Discord or contact our CEO at vasek@e2b.dev if you have any questions.

SDK Bug Fixes & Improvements

• You don’t have to pass API key when starting session if you have the E2B_API_KEY env variable set

• Python SDK now supports both Pydantic v1 and v2

• You can use ~ and relative paths in the filesystem module

• You can set the working directory for the whole playground session via the cwd argument

• You can pass environment variables to the playground via the envVars and env_vars arguments for each SDK respectively

• Users can now use sudo in the playground properly

• Fixed user permissions inside the playground

• Fixed .bashrc/.profile not being loaded inside the playground

• Fixed the incorrect user’s home directory in the playground

This week we released a new version of our SDK, added examples of use cases, and continued working on the infrastructure and custom environments.

New methods

1. Write bytes

   • Writing bytes to a file in our playground allows you to upload binary files like images, audio files, or video files

2. Read bytes

   • Reading bytes from a file inside our playground allows you to download binary files like images, audio files, or video files

Added examples

We added runnable code snippets for the use case examples for AI agents to our documentation:

1. Executing code

2. Cloning GitHub repo & installing deps

3. Running shell commands

All use cases are available in Python and JS.

We continued working on the E2B infrastructure and made several smaller improvements in our docs. If you want to stay in touch, please join our Discord, where we publish updates in real time.

Documentation updates

• You can now run code samples in Python and JS directly in the docs 

  
  

• We added an example of how to run multiple processes with the SDK to our docs

• We made our docs easier to use by highlighting relevant parts

New insights about agents

• We closed the first part of our agent developers interview series with a conversation about building tools for LLM agents with the CEO at Lindy AI

  
  

This week we updated our infrastructure and SDK and worked on the E2B docs.

Infrastructure

• Worked on the custom playgrounds and CLI for managing the AI playgrounds

Updates in SDK

• Improved speed of starting playgrounds in the Python SDK

• Allowed users to set timeouts when using playgrounds in JS SDK to limit the maximum time playground methods can run

• Added logging to the JS SDK

Documentation

• Worked on improving user experience in the documentation, so users can quickly try E2B SDK right from their browser

This week we launched AI Playgrounds, supporting Python and JS/TS. 

It is a sandboxed cloud environment where an AI agent or copilot can freely run code, use a filesystem, use a headless browser, and more.

Try the AI playgrounds here: https://e2b.dev/docs.

Every playground has its own URL, which allows your agent for example to start a web app. AI playgrounds are especially useful for coding copilots and agents. They can easily run linters, formatters, or debug code and feed all this information back to LLM to make the next iteration better.

Quality of life improvements

• An API key is now required for our SDK (get your API key here)

• Added timeouts to asynchronous methods in our SDK

• Improved Python SDK debug logging

• Improved internal overview of the usage of our infrastructure

• Added periodic end to end test of our infra via calling SDKs

This week, we worked on the following:

SDK

• Improved the reliability of our SDK and fixed bugs regarding thread and WebSocket communication

• Brought both the JavaScript SDK and Python SDK to parity

• We improved our release process in our monorepo

Documentation

• Added an option to retrieve API keys

• Added more content based on user feedback

UX improvement

Docs, changelog, and blog are now accessible through e2b.dev/docs, e2b.dev/blog, and e2b.dev/changelog instead of subdomains. 

Community

We talked with the founders of Superagent and Aomni about LLM agents' orchestration, debugging, tracing, and monitoring. Read the published interviews on our blog.

We hosted a community event in Prague where local startups talked about developer tools and AI.

We released new documentation for the E2B SDK that helps you with getting started.

With a single line of our SDK, you can give your AI agent a sandboxed cloud environment where your agent can do any of the following (and more!):

• Run any code

• Run any terminal command

• Install dependencies and programs

• Use filesystem

• Upload and download files

• Access the internet

• Start a web server that's accessible from the internet

• Clone git repositories

• Start any process (even long-running such as a database)

The new docs help you with installation and basic usage and examples are to be added soon. We have fixed reliability-related issues in the SDK.

We’ve been iterating with our early users and fixed a bunch of critical bugs.

If you want early access to the SDK, join our Discord and send us a message!

Agent Protocol Documentation

We put the Agent Protocol on a dedicated website with a nice domain and proper documentation. 

The documentation goes deeper into the incentives of the protocol, guides for adoption, and the details of how the protocol works.


Discuss the updates on our Discord, and if you want to appreciate our work, star our repositories on GitHub.

This week, we have for you an early access version of the E2B SDK.

The E2B REST API gives AI agents their own sandboxed cloud environment. Agents can use these environments to run code, terminal commands, long-running tasks like databases, start servers, and upload and download files.

You can run pip install e2b to get our SDK.

The E2B SDK is available in Python and we are finalizing a JavaScript/TypeScript version. See the Python SDK also in PyPi or go through the examples.

We recently tried out our API in the e2b plugin for ChatGPT which you can check out if you have developer access to ChatGPT plugins.

Next on our roadmap

• Make the API generally available

• Integrate the API and the protocol into popular agent-building frameworks

Contributors our welcome, so in case you are interested, just do a pull request on our GitHub, or discuss more info on our Discord.

The e2b ChatGPT Plugin gives your ChatGPT access to a full cloud environment that is sandboxed and running on e2b infrastructure.

We’ve built these cloud environments to be used specifically by AI Agents. With the plugin, you can accomplish for instance deploying websites, managing databases, executing shell commands, or installing programs via the terminal.

Try out the Plugin here and tell us how you liked it via Discord or by tagging @e2b_dev on Twitter. Contributors and beta testers are welcome.

Agent Protocol Update

We did an overhaul of the Agent Protocol repository. Main updates include:

• Better communication of incentives, roadmap, and usage

• RFC template

• Better code examples 

• New JS SDK

• New agent-protocol CLI to check compliance with the spec.

The protocol serves as a common interface for interacting with any agent. It is techstack-agnostic and open-source.

We're now a few months into building e2b, which means we will be sharing weekly updates. Here's some news we think you'll love in this release.

Get your own instance of smol developer

This week, we have launched a smol developer - your own personal junior developer which you can deploy via e2b. You can connect the developer with three simple steps.

Let us know how you like the smol developer.

Agent Protocol

We have made first steps in developing the Agent Protocol.

The protocol defines an interface for interacting with your agent. Adopting the protocol makes your agent automatically deployable to the e2b platform.

Our goal is to make this protocol open and usable in the whole agent field. This is a collaborative project. We're working on it with the most popular open-source agents, e.g., Auto-GPT, smol developer, or babyagi.

Objectives:

• Common endpoints to communicate with an agent - "start", "step", "stop" for starters. More will come, especially around debugging, tracing, and monitoring

• Tech-stack agnostic - no specific framework required - you can adopt this without using the rest of SDK

• Great DX - easy to adopt and implement - no need to rewrite your existing agent

• Open - any agent builder can implement this and instantly become compatible with any platform supporting this protocol

If you like the protocol, please appreciate the GitHub repo with a star. Contributors are welcome.

Smol Developer Update

With more than 1500 smol developer pull requests, we hit OpenAI API rate limits. We consequently added an option to pass your own OpenAI key in case e2b is hitting OpenAI's rate limits.

Additionally, you can now select which OpenAI model you want to use with smol developer when using your own OpenAI key. Try the updated smol developer here.