Getting started with SERA in Claude Code

2026-01-27 · Source: Ai2 · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure · Depth: Intermediate, quick

Summary

AI2 has released a demo showcasing how to self-host their softverified efficient repository agents (SERA) model using Modal and interact with it via Cloud Code. The process involves installing the Modal and AI2 SERA CLI tools, then setting up Modal to manage cloud GPU resources. The SERA-Modal command initiates a new Modal app, downloads the SERA model from Hugging Face, and launches VLM to host the model with OpenAI-compatible endpoints. This setup allows users to interact with SERA through Cloud Code, demonstrating tasks like updating the deployment to use two GPUs and setting tensor parallelism to two for faster inference. Upon exiting Cloud Code, the system automatically cleans up Modal resources to prevent idle GPU charges.

Key takeaway

For MLOps Engineers or AI Engineers evaluating model deployment strategies, self-hosting the AI2 SERA model on Modal offers a streamlined setup with automatic resource management. Your team can leverage Modal's free credits to experiment with SERA for approximately six to seven hours, gaining practical experience before committing to a specific cloud GPU provider or on-premise deployment. This approach minimizes idle resource costs and provides full control over the model's environment.

Key insights

Self-hosting AI2's SERA model on Modal enables interactive development with Cloud Code.

Principles

• Modal simplifies cloud GPU deployment.
• VLM provides OpenAI-compatible endpoints.

Method

Install Modal and SERA CLI, set up Modal, then use `SERA-Modal` command to deploy the model, interact via Cloud Code, and automatically clean up resources upon exit.

In practice

• Deploy SERA on Modal for free trial.
• Adapt CLI for other cloud GPU providers.
• Monitor SERA's verbose thinking outputs.

Topics

• SERA Model
• Modal Deployment
• Claude Code
• vLLM
• GPU Acceleration

Best for: MLOps Engineer, Machine Learning Engineer, AI Engineer

Related on AIssential

• Build Small with Modal — Modal offers purpose-built, developer-friendly infrastructure for scaling data and compute-intensive AI workloads with fast GPU orchestration and sandboxing.
• Build Your Own Cloud Lab and AI-Powered AWS Agent for $0 Using EC2, Docker, and Claude Code — Build a free, AI-powered AWS cloud lab to manage infrastructure with natural language commands.
• 🧠 Community Wisdom: What to do when non-PMs start shipping directly to production, thoughts on Claude Code’s pricing A/B test, the use of gen AI in games, and more — Community discussions offer practical strategies for LLM optimization and production deployment.
• Anthropic Claude launches on Microsoft Azure Foundry — Claude's Azure deployment on Nvidia hardware expands enterprise AI capabilities and solidifies strategic partnerships.
• Build an AI app with FastAPI and Docker - Coding Tutorial with Tips — Integrate Hugging Face models with FastAPI and Docker for efficient, production-ready ML applications.
• Building Needle: Vibe automation, RAG workflows, and founder lessons — Needle simplifies AI automation, especially RAG and voice AI, through a visual, low-code platform.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Ai2.