Environments
Overview
An environment determines where your assistant runs. By default, assistants run in Vellum Cloud, our managed platform, so you can sign up and go without managing any infrastructure. If you'd rather host the runtime yourself, you can run it locally on your Mac or deploy it to your own GCP, AWS, or custom Linux host. The environment you select affects latency, availability, resource limits, and how much control you have over the underlying infrastructure.
For self-hosted runtimes, you can specify the environment during hatch using the --remote flag:
vellum hatch --remote <local | gcp | aws | custom>Architecture
The following diagram shows how the different environments relate to channels and external providers:
Vellum Cloud (recommended)
Run your assistant on Vellum's managed platform. No cloud accounts or server management required, just sign up and go.
Vellum handles provisioning, upgrades, scaling, and infrastructure so you can focus on using your assistant. Managed assistants use Anthropic (Claude) as the default provider, billing is handled through your Vellum account, and your workspace is encrypted and isolated to you.
- Pros
- Zero setup, always-on, automatic upgrades, no infrastructure to manage. Reachable from web, desktop, mobile, voice, and chat channels.
- Cons
- No direct access to local files or tools on your machine without the desktop app. Provider selection is managed by Vellum.
Local
Run the assistant runtime on the same machine as the desktop app. Useful for development, testing, and privacy-sensitive use cases where you want everything on hardware you own.
vellum hatchWhen running locally, the assistant daemon and gateway both start on your machine. Latency is low and your assistant has direct access to local files and tools.
- Pros
- Low latency, full access to local files and tools, runs entirely on your hardware.
- Cons
- Tied to your machine being on. Uses local compute resources.
User Hosted
Run the assistant on infrastructure you control. This is useful when you need the assistant to stay running independently of your local machine, when you need more compute resources, or when you have specific compliance requirements. Three hosting options are supported:
GCP
Provisions a Google Cloud Compute Engine VM and bootstraps the assistant runtime on it.
vellum hatch --remote gcpRequires gcloud authentication and the GCP_PROJECT and GCP_DEFAULT_ZONE environment variables.
AWS
Provisions an AWS EC2 instance and bootstraps the assistant runtime on it.
vellum hatch --remote awsRequires AWS credentials configured via the standard AWS CLI authentication flow.
Custom
Deploy the assistant to any machine you can SSH into. Set the VELLUM_CUSTOM_HOST environment variable to your target host.
VELLUM_CUSTOM_HOST=user@hostname vellum hatch --remote customThis option gives you full flexibility. Use any Linux machine (on-premises, a VPS, or a VM from any cloud provider) as the assistant's runtime environment.
Choosing an Environment
| Environment | Best For | Requires |
|---|---|---|
| Vellum Cloud | Most users. Zero-ops managed hosting, always-on, accessible from anywhere. | Vellum account |
| Local | Personal use, development, testing, privacy-sensitive workflows | Desktop app, vellum CLI |
| GCP | Always-on assistant on your own infrastructure | GCP account, gcloud CLI |
| AWS | Always-on assistant, AWS-native teams | AWS account, AWS CLI |
| Custom | On-premises, custom infra, any SSH host | SSH access to target machine |