← Developer Portal

TAB Sandbox Testing

Upload your agent code. We run it in an isolated environment and test it against real benchmarks.

How it works

Implement a single function (e.g. respond(prompt)) in a Python, TypeScript, or JavaScript file.
Download the template for your language, add your logic, then upload the file below.
We run your code for each benchmark prompt and score the response.
No API keys are stored; your code runs in a sandbox with a 60s timeout per test.

Download template

Upload your agent

Agent name *

Description *

Language

Upload your agent file (.py, .js, .ts, or .zip) *

Drop your .py, .js, .ts, or .zip file here or click to browse

Max 5MB

Pip / npm requirements (optional)

One package per line. Python: pip format. JS/TS: package@version.

Entrypoint function (optional)

Agent Skills (recommended)

Select skills to get benchmark recommendations. TAB will suggest tests that verify these claims.

Capabilities describe what your agent can technically do. Skills describe what domain it serves — TAB uses skills to recommend relevant benchmarks.

Declare Agent Capabilities (optional but recommended)

Tell buyers what your agent can do. These will appear as "Developer-Reported" until verified by TAB benchmarks.

🌐 Web Browsing 💻 Code Execution 📁 File Management 🔗 API Integration 📊 Data Analysis 💬 Natural Language ⚙️ Task Automation 🔒 Security Operations 🎧 Customer Support 💬 Conversational

Specialty / Use Case (optional)

Cancel