ramalama/Roadmap.md

RamaLama Roadmap

This document outlines upcoming features and initiatives for RamaLama.
Contributors can comment on or contribute to the issues linked here.

---

🚀 New Commands

• ramalama summarize
• ramalama audio2text
• ramalama prompt2img sd <model>
• ramalama --container-name=<name> upload <file>
  • Upload files (image, text, PDF, etc.) to a containerized model.
  • No action initially; waits for a prompt such as “summarize this text.”

---

🛠️ Runtime & Design Improvements

• Make MODEL runtimes more pluggable
  Define a syntax that simplifies adding new runtimes.
  Current runtimes:

  • llama.cpp
  • whisper.cpp
  • vllm
  • stable-diffusion
  • OpenVINO

• OpenVINO integration

  • Coordinating with Intel for CPU/accelerator support.

• Model OCI Artifact support

  • Podman 5.6: podman-remote artifact support.
  • Match Docker behaviour for storing models as OCI artifacts.
  • Reference: CNCF sandbox issue #358.
  • Evaluate defaulting --container mode to automatically convert models to OCI artifacts.

---

📚 Retrieval-Augmented Generation (RAG)

• Add support for RAG pipelines.
• Explore MCP (Model Context Protocol) integration.

---

🐑 Llama-stack Features

• Consolidation of images.
• Default AMD RamaLama images to:
  • quay.io/ramalama/ramalama
• Assess Intel-specific image defaults.

---

🎮 Vulkan Efforts

• Expand support for Vulkan backends.
• Ensure compatibility across GPUs.

---

🧩 Additional Tooling

• VSCode Plugin

  • Start a RamaLama container with the local project mounted.
  • AI-assisted code analysis and suggestions (similar to GitHub Copilot).
  • Key difference: data stays on the developer’s machine.
  • May require MCP server integration.

• Automated Image Detection

  • Detect best base image from a compatibility matrix.
  • Automatically select and execute commands with the correct image.

---