ClickUp is the everything app for work, software built to replace all software. Projects, Docs, Chat, dashboards, agents, and AI all live in one platform, trusted by 85% of the Fortune 500. Because a company's work lives in one system, ClickUp has something standalone AI tools never will have: complete context.
Brain² is the AI built with that context. It's your company's Brain: multiplayer, self-improving, and running every frontier model with full knowledge of your work. Brain² doesn't just answer questions, it does the work. Ask for a report, a deck, or a dashboard, and Brain² inspects the files, writes and runs code against the real data, and returns a finished artifact right in the conversation.
That code touches sensitive workspace data, has to start fast enough to feel native mid-conversation, and has to scale to millions of users. ClickUp built that execution layer with E2B.
“We see code execution as a fundamental primitive of agentic action in the future, and working with E2B has been absolutely fantastic to bring that to market.”
The challenge
Large files break the context window
Brain² already understands the work inside ClickUp: tasks, docs, chats, and the relationships between them. But finishing that work often takes code. Analyzing a year of sales data means several large files, far more raw data than a model should read directly. Loading every row into the context window also forces truncation, so the answer reflects only a fraction of the file.
Code written by a model needs a hard isolation boundary
ClickUp serves enterprises in regulated industries like healthcare and financial services, so the systems that touch their data have to meet SOC 2 and HIPAA requirements. Brain² works with uploaded files and workspace data, and every execution has to respect the same permissions and data boundaries as the rest of ClickUp.
For the execution layer, that means isolation between customers, users, and conversations. Code running for one request must not be able to reach another sandbox, a neighboring workload, or ClickUp's production systems. When the work is done, the environment has to disappear and keep user data secure.
Demand spikes at 9 a.m.
Millions of users start work at roughly the same time, and Brain²'s sandbox requests spike with them. Teams schedule automations, standups, and recurring reports for the start of the workday, so many workflows use Brain²'s code execution capability at once. The infrastructure has to absorb that burst without queuing requests or asking ClickUp to plan capacity around it.
The solution
From a request to a finished artifact
Users invoke Brain² from wherever they are already working: a task, a doc, a chat thread, or the sidebar. Brain² reads the request and decides which tools it needs. When a request calls for data analysis, file processing, or a long chain of tool calls, Brain² provisions an E2B sandbox, uploads the files, generates and runs code inside it, and returns the chart, report, or presentation directly in the conversation. The files stay in the sandbox instead of filling the context window, so Brain² works through the full dataset and the model only sees the computed result.
A user can even upload a screen recording and ask Brain² about it. For example, a user uploads a product walkthrough and asks for documentation. Inside the sandbox, Brain² inspects the video frame by frame, transcribes the audio, extracts the key steps, and creates a document with the documentation.
“We can upload all those files into the sandbox, and then the model is only responsible for generating code to analyze them, instead of having to load all file content into its context. Theoretically speaking, with this approach you can 10x or even 20x the amount of context that you're looking at in any one request.”
A full computer with 68 pre-installed tools
ClickUp uses E2B templates to define the environment before a request arrives. Brain²'s base template ships with 68 tools covering data analysis, visualization, document and spreadsheet processing, media analysis, and report generation. Purpose-built templates add dependencies for specialized workloads, which supports workflows across engineering, sales, finance, leadership, marketing, and GTM.
Preloading the environment removes package installation from every request, so Brain² starts executing right away and users get answers faster.
Isolated, short-lived, and resumable
Every sandbox runs in its own E2B microVM, with its own kernel and execution environment, so that code running for one conversation cannot reach another. E2B met ClickUp's compliance requirements from the start, including SOC 2 and HIPAA.
Isolation protects the data while the sandbox is running. Most sandboxes live for 15 to 30 minutes and serve a single request. When the conversation ends, the sandbox is terminated and its contents are deleted, so user data is protected.
Some workflows run longer, with several messages or repeated edits to an output. For those, ClickUp uses E2B's pause-and-resume feature to keep the execution state between requests, so Brain² returns to the same environment instead of rebuilding it.
“At the end of the day, any infrastructure behind ClickUp Brain² has to pass the same enterprise security review as the rest of ClickUp. E2B clears that bar at every layer, from SOC 2 and HIPAA requirements down to isolated microVMs where every sandbox has its own kernel.”
Building in-house would have taken ten times as long
ClickUp considered building the sandbox infrastructure in-house. That would have meant standing up a new team to build and maintain its own microVM platform, and keeping engineers on it instead of on Brain². However, the AI team had already tested E2B during an internal hackathon, where the SDK was easy to pick up, the documentation was clear, and sandboxes started quickly. The build-versus-buy decision was easy.
“Going down the E2B route allowed us to get something out to our users within the space of maybe a week or two. Whereas if we had built it in-house, it probably would have been 10x that duration.”
The results
With E2B, ClickUp:
- Shipped sandboxed code execution in 1-2 weeks, roughly 10x faster than building it in-house
- Spins up thousands of sandboxes within a minute to absorb demand spikes
- Lets Brain² analyze files and datasets far larger than any model context window
For the user, the experience is a request in a ClickUp conversation and a finished result in the same place. Behind it, Brain² starts an isolated sandbox from a prebuilt template, a full Linux computer with its own filesystem and 68 preloaded tools, and low-latency startup keeps the whole workflow moving at conversation speed.
Looking ahead
As Brain² takes on more features and use cases, ClickUp expects to run more of its work in E2B sandboxes. Agents get more capable when they can write their own solutions. If an agent has no built-in way to achieve something, it can write the code to do it inside a sandbox. Every new capability Brain² gains widens the landscape of what agents can build for themselves.
“It's consistently surprising when you give a frontier model access to a whole computer. You tell it it can do whatever it wants, it goes off, and it will blow your mind. We're still discovering the applications every day.”