Agent as a Service A Chutes.ai project
Isolated sandboxes for running Agents and LLM generated code. Safe. Scaleable. Affordable.

Launch isolated AI agents in one API call. YOLO speed, sandbox safety.

Agent as a Service wraps Sandy’s sandbox infrastructure into a ready agent runtime. Start Claude Code, Codex, or custom agents without building orchestration, and let them execute end-to-end with confidence.

Request access See the platform
1 API call to launch
2s Agent boot time
∞ Safe parallel agents
Live in minutes

Pick an agent profile, drop in your API key, and stream outputs securely.

Agent as a Service workflow

Why Agent as a Service

Autonomous agents need isolation to scale.

Agent as a Service lets you run YOLO agents without trading away security.

Built on Sandbox as a Service, the Chutes project that provides sandbox infrastructure for agents.

YOLO mode unlocks speed

Agents are most effective when they can run end-to-end without constant approval prompts. We prewire the runtime so they can move fast.

Isolation makes autonomy safe

Each agent gets a hardened Sandy sandbox, with tight resource limits and audit trails. You can run large fleets without compromising your core systems.

Prefer pure sandboxes? Visit Sandbox as a Service →

Platform

Prewired agent runtime on top of Sandy.

Agent as a Service packages best-in-class agent tooling into hardened sandboxes so you can focus on outcomes, not orchestration.

Agent launch in seconds

Boot Claude Code, Codex, or custom agents with a single API call and a ready runtime.

  • Agent profiles with preinstalled tools
  • Streaming logs + exec output
  • Fast terminate + cleanup

Stateful workspace mounts

Persist files between agent runs or start fresh every time. You stay in control.

  • Optional workspace snapshots
  • Secure artifact export
  • Configurable cache budgets

Safety rails built in

Every run is isolated, audited, and limited by default to keep your systems safe.

  • Per-agent policy controls
  • Network egress guards
  • Audit-ready logs

Get running in 5 minutes

  1. Create your Agent API key
  2. Start an agent sandbox with one request
  3. Send tasks, stream output, terminate

Works with existing LLM pipelines and internal tools.

Agent launch 2 calls 
curl -X POST https://api.agent-as-a-service.online/v1/agents \
  -H \"Authorization: Bearer $AGENT_API_KEY\" \
  -d '{\"profile\": \"claude-code\"}'

curl -X POST https://api.agent-as-a-service.online/v1/agents/ag_72b4/tasks \
  -H \"Authorization: Bearer $AGENT_API_KEY\" \
  -d '{\"prompt\": \"Refactor the repo and open a PR\"}'
Inference included

Bring your own inference via Sign In With Chutes. Auth and routing are handled, so agents can run instantly without extra wiring.

API docs

Orchestrate agents with a few requests.

Authenticate once, launch agents, stream output, and keep full control.

Authentication

Send your Agent API key as a bearer token on every request.

Authorization: Bearer $AGENT_API_KEY

Core endpoints

  • POST /v1/agents
  • POST /v1/agents/:id/tasks
  • GET /v1/agents/:id/events
  • POST /v1/agents/:id/terminate

Lifecycle

  1. Launch an agent with a profile and tool access
  2. Send tasks, stream logs, pull artifacts
  3. Terminate the agent sandbox and clean up

Advantage

Massive cost advantage, predictable performance.

Agent as a Service runs on dedicated Sandy infrastructure. We pass the savings directly to teams that need reliable agent sandboxes at scale.

Up to 95% lower sandbox cost

From $5 per sandbox / month for entry usage.

Cost leader

Cost reality check

20 always-on sandboxes, monthly estimates:

E2B
$1,918
Daytona
$1,918
Northflank
$720
Modal
$2,740
Cloudflare
$1,051
Agent as a Service
$100

Price comparison uses 20 sandboxes for one month across providers.

What you get

  • Dedicated hardware, no noisy neighbors
  • Low-latency SSD + cached dependencies
  • Transparent scaling knobs
  • Run with or without Docker socket
  • Telemetry snapshots + event history
  • Premium support with SLA options

Pricing

Flexible plans built for teams of every size.

Start small, scale into multi-agent fleets, and keep every run under control.

Starter

From $5/month

Smallest config: 0.5 vCPU / 1 GB RAM. 1 vCPU / 2 GB RAM is $20.

  • Up to 20 concurrent sandboxes
  • Agent API + telemetry
  • Community support
Get started
Most popular

Scale

Custom/month

  • Dedicated agent node
  • Custom sandbox sizes
  • Priority support + SLAs
Talk to sales

Enterprise

Let’s talk

  • Multi-region clusters
  • Private network peering
  • Custom compliance requirements
Book a call

Security

Isolation-first, designed for production workloads.

Every sandbox is locked down by default. You can opt into elevated permissions only when you need them.

Hardened by default

CAP drop, no-new-privileges, and strict resource controls.

Zero trust networking

Outbound controls, per-sandbox host suffixes, and private ingress routing.

Full audit trail

Telemetry events stored in Postgres with export-ready history.

FAQ

Questions teams ask before switching.

How fast can I start a sandbox?

Cold start is typically 2-4 seconds. Prewarming can reduce this further.

Do you support Playwright and browsers?

Yes. The runtime ships with optional Playwright + Chromium builds.

Can we bring our own Docker images?

Yes. Enterprise plans support custom runtime images and internal registries.

Launch your agent fleet today.

Ready to let agents run end-to-end without building the infrastructure?

Back to top