Deploy background agents
with one command
The deploy and runtime layer for AI agents. Point Castari at any project and ship every agent to an isolated cloud sandbox with one command.
npm install -g @castari/cli
Deploy agents in seconds using the Claude Agent SDK
Agents to production in seconds using
Highest security
Every agent runs in a fully isolated sandbox with zero shared state. Network policies, ephemeral environments, and end-to-end encryption keep your data and your users protected at every layer.
Unlimited customization
Bring your own models, tools, and workflows. Castari gives you full control over runtime configuration, MCP integrations, and deployment parameters — no abstractions standing between you and your agent.
Infinite scale
From a single request to thousands of concurrent agents, infrastructure scales automatically to match demand. No capacity planning, no cold starts, no limits on what you can build.
3 steps to deployment
Import Your Project
Import a Claude Agent SDK project from GitHub or generate a new agent from a prompt or template.
One click deploy
Click deploy or run cast deploy. Your agent runs in an auto-scaling sandbox with built-in logging and observability.
Iterate on your agent
View tool traces, inspect output logs, and refine your agent using real production data. Infrastructure scales automatically.
- IMPORT
- DEPLOY
- ITERATE
Import Your Project
Import a Claude Agent SDK project from GitHub or generate a new agent from a prompt or template.
One click deploy
Click deploy or run cast deploy. Your agent runs in an auto-scaling sandbox with built-in logging and observability.
Iterate on your agent
View tool traces, inspect output logs, and refine your agent using real production data. Infrastructure scales automatically.
Features
Auto-scaling sandboxes
Run agents in secure sandboxes designed for long-running workloads. Infrastructure scales automatically based on demand.
MCP gateway
Connect agents to any external service through a unified gateway. Route to Slack, GitHub, Gmail, and more with zero configuration.
Model router
Intelligently route requests across multiple AI providers. Optimize for cost, latency, and capability across your agent fleet.
Agent observability
Full visibility into agent behavior with tool traces, output logs, and real-time monitoring. Debug and iterate with production data.
Auto-scaling sandboxes
Run agents in secure sandboxes designed for long-running workloads. Infrastructure scales automatically based on demand.
Model router
Intelligently route requests across multiple AI providers. Optimize for cost, latency, and capability across your agent fleet.
MCP gateway
Connect agents to any external service through a unified gateway. Route to Slack, GitHub, Gmail, and more with zero configuration.
Agent observability
Full visibility into agent behavior with tool traces, output logs, and real-time monitoring. Debug and iterate with production data.
Agents can use the same tools as humans
Give your agents access to browsers, terminals, APIs, and databases — the same interfaces your team already uses. No custom integrations needed. They work alongside your existing workflows, not around them.
Your agents are limitless.
Research agents
Dev agents
Finance agents
DevOps agents
Data agents
Customer support agents
FAQs
Castari is a deployment platform for Claude Agent SDK agents. Import your project, run cast deploy, and your agent is live with auto-scaling, logging, and observability built in.
No. Castari works with standard Claude Agent SDK code without modification. You keep full ownership of your codebase with zero vendor lock-in—if you ever want to self-host, your agents will run anywhere. We simply provide the infrastructure layer.
Deploying agents yourself means managing containers, configuring secure environments, handling scaling logic, and monitoring execution. Castari handles all of this out of the box with production-grade infrastructure and no DevOps overhead.
Every agent runs in an isolated sandbox environment. We use a variety of sandbox providers to ensure security and performance. The ability to select your own sandbox provider is coming soon.
Castari is free during our public beta. When we move to general availability, we'll use usage-based pricing—you'll pay only for the compute your agents use, not per seat or per agent.
Castari supports any model with the Claude Agent SDK—not just Claude. Use OpenAI, xAI, open source models, or any provider you prefer. Your agent code stays the same.
Getting started takes less than 5 minutes. Install the CLI with npm install -g @castari/cli, authenticate with cast login, and deploy your Claude Agent SDK agent using cast deploy. Your agent will immediately be accessible via REST API or our SDK.
