What is the Crusoe Agent Platform?
The Crusoe Agent Platform is a cloud for building and running AI agents. You upload a folder of Python code, and the platform turns it into a live HTTPS endpoint you can chat with — no servers to set up, no containers to write, no infrastructure to babysit. This page explains what the platform is, shows you the map, and points you to your first five minutes.
Who it's for
- Developers who want to ship an AI agent, a function, or a web service without learning Kubernetes.
- Teams that need shared projects, roles, API keys, quotas, and an audit trail around those workloads.
- Agent builders using ADK, LangGraph, or CrewAI — the platform runs your unmodified framework code.
You interact with the platform two ways:
- The web console at https://console.codyhill.dev — sign in, pick a project, and manage everything from your browser.
- The CLI,
platformctl— the same power in your terminal, for scripting and CI. See Install the CLI.
The 30-second mental model
Everything you create lives in a project (your team's private workspace). You deploy code through the management plane (the API the console and CLI talk to, always authenticated). Your deployed agent then serves requests on the data plane, and the platform wires it to managed services automatically: conversation history, long-term memory, and a code sandbox.
When nobody is talking to your agent, it scales down to zero running instances — it costs nothing to sit idle — and wakes up automatically on the next request. That idea, called scale-to-zero, runs through the whole platform. Read more in Core concepts.
Every service at a glance
| Service | What it does | Docs |
|---|---|---|
| Agent Engine | Deploy a folder of ADK, LangGraph, or CrewAI agent code as a live chat endpoint with sessions, memory, and tools. | Overview |
| Serverless | Run containerized web services that scale with traffic, down to zero. | Overview |
| Functions | Deploy a single handle(event) file (Python, Node.js, Go, or Ruby) as an HTTPS endpoint. | Overview |
| VectorDB | Managed vector indexes for similarity search — the storage behind agent long-term memory. | Overview |
| MemoryStore | Managed key-value stores — the storage behind agent conversation sessions. | Overview |
| Pub/Sub | Topics and subscriptions for sending messages between services. | Overview |
| Code Sandbox | Run untrusted Python in an isolated, single-use environment — powers the run_python agent tool. | Overview |
| Durable Execution | Long-running workflows that survive restarts and retry failed steps. | Overview |
| Secrets Manager | Store credentials once, inject them into workloads as environment variables. | Overview |
| MCP Servers | Publish Python tool code as a hosted Model Context Protocol endpoint with immutable versions and rollback. | Overview |
| Console | The web UI: projects, IAM, service accounts, API keys, quotas, audit log, and every service page. | Overview |
| Security & IAM | Organizations, projects, the two-axis role model, API authentication, and break-glass auditing. | Overview |
Coming from another cloud? The service mapping page translates AWS, GCP, and Azure product names into ours, honestly — including where we have no equivalent yet.
How it compares
| If you use... | The closest thing here | Honest difference |
|---|---|---|
| GCP Vertex AI Agent Engine | Agent Engine | We run the same open-source ADK. We don't have GCP's regions, SLAs, or enterprise compliance programs. |
| AWS Bedrock AgentCore | Agent Engine + Code Sandbox | One product, one deploy path — no "Classic vs. v2" split. AWS has far more managed model choices. |
| Azure Foundry Agent Service | Agent Engine | We are code-first only; there is no no-code "prompt agent" builder. |
| AWS Lambda / GCP Cloud Run | Functions / Serverless | Same scale-to-zero idea. We support 4 function runtimes; the big clouds support more, in more regions. |
Honest alpha status
This platform is in alpha. We would rather tell you the rough edges up front than let you find them at 2 a.m.:
- No self-service sign-up. Accounts are created by an administrator or through an invitation link. See Create an account.
- No public API hostname yet. The web console is the public front door. Scripts and the CLI reach the API through an admin-provided endpoint, the
$CAI_APIenvironment variable, or an automatic port-forward. See Install the CLI. - The CLI is built from source. There are no prebuilt
platformctldownloads yet. - Sign-in sessions last 12 hours and cannot be revoked early. There is no multi-factor authentication yet.
- No SLA, no uptime guarantee, no multi-region. Don't run your pacemaker on it.
- The full list lives in Known issues.
What alpha does not mean: your data is still isolated per project, every state change is audited, platform staff cannot read your project's resources without a visible, expiring break-glass grant, API keys are shown exactly once and stored only as hashes, and secret values are write-only — the API can never return them.
Start here
- Create an account — get signed in and pick your project.
- Install the CLI — optional but worth it.
- Deploy your first agent — the five-minute flagship quickstart.
- Deploy your first function — a hello-world HTTPS endpoint.
- Core concepts — how projects, roles, keys, and quotas fit together.