The Execution platform Bulit for AI Agents
Runloop gives your agents a full development environment -- isolated, stateful, and fast enough to run at production scale. Every primitive is API-first, so you control the lifecycle from first boot to final snapshot.
Runloop is two products in one platform. This page covers Execution -- the infrastructure where your agents run. The Evaluation & Benchmarking page covers how you measure whether they are improving. No other sandbox provider offers both.
Secure, isolated environments where your agents do real work
Devboxes are full Linux virtual machines -- not containers, not serverless functions. Each one runs on a dedicated MicroVM with hardware-level isolation, giving your agent the same capabilities a human developer has: filesystem access, shell execution, networking, and persistent state.
Define once. Launch thousands.
Blueprints are programmable templates for Devbox environments. Define your system packages, language runtimes, dependencies, and file mounts in code, and every Devbox launched from that Blueprint starts fully configured. Docker layer caching means repeated launches skip setup entirely.
Pause, branch, and resume agent work
Snapshots capture the full disk state of a running Devbox. Resume from a snapshot to pick up exactly where you left off, or fork it into multiple parallel Devboxes to explore different approaches simultaneously.
Your agents use credentials without ever holding them
The most common vector for credential compromise in agent systems is prompt injection. An adversarial input tricks the agent into revealing API keys. Runloop Credential Gateway eliminates this attack surface. Real credentials never exist on the Devbox.
One endpoint for all agent tools, with per-tool permissions
MCP Hub aggregates all tool servers behind a single endpoint with fine-grained control over which tools each agent can access. Your agent connects to one URL, sees a unified tool list filtered by its permissions, and calls tools by name. Restricted tools are invisible -- not denied, invisible.
Control what your agents can reach
Fine-grained control over network access at the Devbox level. Network Policies define DNS-level allowlists -- everything not explicitly permitted is blocked. Policies are lifecycle-aware: broad during builds, locked during runtime, fully restricted during snapshot capture.
The workflow surface for building, debugging, and shipping agents
Four tools for the full agent development lifecycle -- from first prototype to production deployment.
Three lines to launch. Full lifecycle control from there.
Every execution primitive -- Devboxes, Blueprints, Snapshots, Credential Gateway, MCP Hub, Network Policies -- is accessible through the same SDK.
import runloop
# Launch a Devbox with credential protection and tool access control
devbox = runloop.devboxes.create(
blueprint_id="bp_agent_runtime",
credentials=[
{"name": "RL_ANTHROPIC", "secret": "anthropic-api-key"},
{"name": "RL_GITHUB", "secret": "github-pat-readonly"}
],
mcp_config="github-readonly",
network_policy="runtime-locked"
)
# Execute a command
result = runloop.devboxes.execute_sync(devbox.id, "python run_agent.py")
# Snapshot for later resume or parallel branching
snapshot = runloop.devboxes.snapshot(devbox.id, name="checkpoint-1")
import Runloop from 'runloop';
const devbox = await runloop.devboxes.create({
blueprintId: 'bp_agent_runtime',
credentials: [
{ name: 'RL_ANTHROPIC', secret: 'anthropic-api-key' },
{ name: 'RL_GITHUB', secret: 'github-pat-readonly' }
],
mcpConfig: 'github-readonly',
networkPolicy: 'runtime-locked'
});
const result = await runloop.devboxes.executeSync(devbox.id, 'node run_agent.js');
const snapshot = await runloop.devboxes.snapshot(devbox.id, { name: 'checkpoint-1' });
# Install the Runloop CLI
curl -fsSL https://runloop.dev/install | sh
# Authenticate
runloop auth login
# Create a sandbox
runloop sandbox create --image python:3.12 --cpu 2
# Execute a command
runloop sandbox exec --id sb_abc123 "python -c 'print(42)'"
# List active environments
runloop sandbox list --status running
# Attach credentials
runloop credentials attach OPENAI_API_KEY --sandbox sb_abc123Built for production from day one
Enterprise security and compliance across every deployment model.
Execution is half the platform. Evaluation & Benchmarking is the other half -- run public benchmarks, build private evaluation suites, integrate regression testing into CI/CD, and generate fine-tuning signals. All on the same infrastructure.
Start building in minutes
Launch your first Devbox with three lines of code. Every execution primitive is API-first, with credential protection and tool access control built in.
Need enterprise deployment? Talk to Sales
