Stop paying for your agents to sleep.
Persistent, isolated containers for AI agents. They hibernate when idle — costing nothing — and wake in seconds on a webhook, cron job, or API call.
Agents that deploy agents.
Your orchestrator agent creates worker cages, injects tasks via environment variables, collects results, and tears them down. The full lifecycle is available via CLI and REST API — from inside a cage.
No human required. Each sub-agent gets its own isolated environment with persistent storage. When it's done, the orchestrator destroys it and reclaims the credits.
- ✓Full lifecycle via CLI or REST API
- ✓Dynamic env var injection per sub-task
- ✓Automatic cleanup — destroy when done
- ✓Each sub-agent fully isolated in its own cage
Your agent works in bursts. Your bill shouldn't be 24/7.
Most agent workloads are bursty — a webhook fires, the agent handles it, then nothing for hours. LobsterCage hibernates when idle and wakes on demand, so you only pay for the time your agent is actually thinking.
AWS EC2
$60/mo
t3.large (2 vCPU / 8 GB) running 24/7.
Paying for 24/7 compute
LobsterCage
$21.60/mo
Pay only while your agent is active. Hibernate the rest.
Paying for ~45 hours/mo
Hibernate when idle
Cages automatically sleep after a configurable idle timeout. Zero cost while hibernating. State persists.
Wake on demand
Webhooks, cron schedules, or API calls wake your agent in seconds. Cages pre-warm 5 minutes before scheduled jobs.
Keep everything
Files, databases, environment — persistent across every hibernate/wake cycle. Your agent picks up right where it left off.
Use Cases
One agent per project. Not one agent for everything.
When separate runtimes are cheap and easy, you stop cramming every job into one overloaded agent. Give each project its own cage.
Coding Agents
Give Claude, GPT, or any LLM a full Linux environment with git, node, python. Persistent repos and tools survive every hibernate/wake cycle. SSH in to pair program.
Messaging Bots
Deploy Telegram bots that respond 24/7. Wake-on-webhook means you only pay when messages arrive. Each bot gets its own cage and context.
Scheduled Automation
Scrape, summarize, report, deploy. Cron-driven agents that wake on schedule, do the work, and sleep between runs. State persists across every cycle.
Platform
What each cage gets
Persistent Storage
Your agent's files, databases, and state survive restarts and hibernation. EFS-backed volumes mean nothing is lost between sessions — not your SQLite database, not your git repos, not your logs.
Wake-on-Webhook
Message arrives. Cage wakes. Agent responds. Cage sleeps. Telegram, Slack, Discord — any webhook source.
Cron Scheduling
Run jobs on schedule. Cages pre-warm 5 minutes early, execute the task, and hibernate automatically when done.
SSH & Tunnels
Real SSH into your agent's environment. Port-forward local services. Debug live in production.
Full Isolation
Every cage is a dedicated Fargate container. No shared filesystems, no lateral movement, no blast radius. Secrets are AES-256 encrypted.
No Surprise Bills
Prepaid credits with hard spending caps. When credits run out, cages stop — never an unexpected charge on your card.
Pricing
Start free. Scale as you grow.
Only pay when your agent is running. Hibernated cages cost nothing.
Free
Test with one cage
- 500 credits/mo
- 1 cage (Starter size)
- 8 compute hours/mo
- Community support
Starter
Run a few agents in production
- 6,000 credits/mo
- Up to 3 cages
- All instance sizes
- Persistent storage
- 20 crons per cage
Builder
Real multi-agent workflows
- 20,000 credits/mo
- Up to 10 cages
- All instance sizes
- 50 crons per cage
- 200 env vars per cage
- Priority support
Team
Organizations running at scale
- 100,000 credits/mo
- Up to 25 cages
- 100 crons per cage
- Org & team management
- Priority support & SLAs
Credits are consumed only while running — hibernated cages use zero. 1 credit = 1 minute of Starter compute. See full pricing details →
Deploy your first agent in 60 seconds.
Start free. Create a cage from a template. Your agent keeps its files, state, and identity between sessions — and you only pay when it's running.