lamia-cloud

Cloud execution backend for Lamia. Run .lm scripts once with --remote, deploy scheduled cloud jobs, and prepare for upcoming cloud trigger support. Currently supports GCP.

For common agent use cases, you usually do not need to build custom cloud-agent infrastructure from scratch before shipping with Lamia.

Installation

pip install "lamia-lang[cloud]"

Prerequisites

• GCP project with billing enabled
• Application Default Credentials: gcloud auth application-default login

All required GCP APIs (including Service Usage) are enabled automatically on first deploy.

Quick Start

1. Add a cloud section to your project's config.yaml:

cloud:
  provider: gcp
  project_id: my-gcp-project
  location: us-central1  # optional, default: us-central1

2. Run a script once in the cloud with --remote:

lamia my_script.lm --remote

Use this one-shot run to validate cloud execution, permissions, and logs before adding a schedule.

3. Schedule your script with the --remote flag:

lamia schedule add my_script.lm --every day --remote

The --remote flag tells lamia to deploy and run the script in the cloud instead of locally.

Managing Schedules

lamia schedule list              # shows all jobs (local + cloud) with live status
lamia schedule add X --remote    # deploy and schedule a new cloud job
lamia schedule remove <id>       # tears down cloud resources and removes the job

How It Works

1. lamia <script>.lm --remote packages your project and runs it as a Cloud Run Job (one-shot)
2. lamia schedule add <script>.lm --remote deploys the same cloud job with Cloud Scheduler
3. Cloud Scheduler triggers the job on your cron schedule
4. Logs are available in Cloud Logging
5. lamia schedule list fetches live execution status from the cloud

LLM on Cloud — Vertex AI

Scripts that use LLM calls run through Vertex AI on cloud. This gives you:

• No API keys — authentication via IAM, no keys to store, rotate, or leak
• Budget control — Vertex AI quotas and billing alerts
• Secure by default — no API key transport or storage, traffic stays within GCP

Supported Models

Provider	Cloud routing
Anthropic (Claude)	Runs natively on Vertex AI — same models, same quality
Google (Gemini)	Runs natively on Vertex AI
OpenAI (GPT, o-series)	Automatically mapped to Gemini by tier (strong/medium/light) with runtime selection of the best available current Gemini model

Anthropic and Google models run as-is. OpenAI models are mapped because they're not available on Vertex AI — tier classification is stable while the selected Gemini model is discovered dynamically at runtime.

Configuration Reference

Field	Required	Default	Description
cloud.provider	Yes	—	Cloud provider (currently gcp)
cloud.project_id	Yes	—	Your GCP project ID
cloud.location	No	us-central1	Region for Cloud Run deployment

No environment variables are required.

Troubleshooting

• If Vertex AI access is not enabled yet, lamia-cloud logs a project-specific URL and attempts to open it automatically in your browser: https://console.cloud.google.com/vertex-ai?project=<your-project-id>
• After accepting terms, re-run the schedule/install command once.

Development

git clone https://github.com/lamia-lang/lamia-cloud.git
cd lamia-cloud
pip install -e ".[dev]"
pytest

Releasing

git tag v0.1.0
git push origin v0.1.0