Agent Infrastructure
Unified infrastructure primitives for AI agents — sandboxes, browsers, code execution, and file systems — with integrated billing, auth, and observability.
What is Agent Infrastructure?
AI agents need more than just LLM calls. They need to execute code, browse the web, manipulate files, and interact with the real world. Today, wiring up these capabilities means juggling multiple providers, billing accounts, and auth systems.
Transactional provides four core infrastructure primitives purpose-built for AI agents, unified under a single API key, billing account, and observability layer.
The Four Primitives
| Primitive | What It Does | Powered By |
|---|---|---|
| Sandboxes | Isolated execution environments (microVMs) | E2B (Firecracker) |
| Browsers | Headless browser sessions for web interaction | Browserbase (Chromium) |
| Code Execution | Secure multi-language script execution | Piston (self-hosted) |
| File Systems | Persistent file storage for agent workspaces | S3-compatible storage |
Why Transactional?
Building agent infrastructure from scratch means integrating, billing, and monitoring each provider independently. Transactional wraps best-in-class providers into a single developer experience.
| Without Transactional | With Transactional |
|---|---|
| Separate accounts for E2B, Browserbase, S3 | One API key for everything |
| Multiple billing dashboards | Unified billing with per-agent cost breakdown |
| No cross-provider auth | Single auth layer across all primitives |
| Build your own logging | Full observability via Observability |
| Manual integration work | Composable primitives that work together |
Build vs. Buy
We deliberately chose best-in-class providers for each primitive rather than building everything from scratch:
- Sandboxes — E2B's Firecracker microVMs deliver sub-200ms cold starts with battle-tested isolation
- Browsers — Browserbase provides managed Chromium with stealth mode and anti-detection built in
- Code Execution — Piston is self-hosted for full control over language runtimes and resource limits
- File Systems — S3-compatible storage gives S3-compatible storage with zero egress fees
Integration with Existing Modules
Agent primitives are designed to compose with the rest of the Transactional platform:
- AI Gateway — Route LLM calls through the gateway for caching, failover, and cost tracking while your agent uses sandboxes and browsers
- Memory — Persist agent context and conversation history across sessions
- Observability — Every sandbox creation, browser action, and code execution is automatically traced
- Email — Agents can send emails as part of their workflow using the same API
- Auth — Authenticate agent sessions with the same auth infrastructure your users use
The Composition Model
The real power is combining primitives. A single agent can:
- Receive a task via your app
- Spin up a sandbox to install dependencies and run complex scripts
- Open a browser to research information on the web
- Execute quick code snippets to transform data
- Store results in a file system workspace
- Send a summary email to the user
All under one API key, one bill, one trace.
Getting Started
- Quickstart — Deploy your first agent in 5 minutes
- Architecture — Understand how the pieces fit together
- Sandboxes — Isolated execution environments
- Browsers — Headless browser sessions
- Code Execution — Multi-language script execution
- File Systems — Persistent file storage