s-kostyaev/ellama

What it solves

Ellama integrates large language models (LLMs) directly into the Emacs editor, allowing users to perform AI-assisted tasks without leaving their environment or relying on a web application. It provides a unified interface for chat, one-shot commands, and complex agentic workflows within the editor.

How it works

Ellama acts as a frontend for various LLM providers via the llm package. It supports a wide range of backends including Ollama (for local models), OpenAI, Claude, Gemini, Vertex AI, and Llama.cpp. The system manages context by allowing users to attach buffers, files, directories, images, and audio files to requests. For more complex tasks, it implements agent loops that can read project instructions from AGENTS.md, utilize reusable skills, and execute a plan-and-act cycle.

Who it’s for

Emacs users who want to leverage LLMs for coding, writing, and data extraction directly within their text editor, as well as developers who prefer local-first AI assistants with robust context management.

Highlights

• Multi-Provider Support: Compatible with Ollama, OpenAI, Claude, Gemini, and others.
• Agentic Capabilities: Supports plan-and-act loops, tool calls, and project-specific instructions via AGENTS.md.
• ** keluarga Context Management**: First-class support for adding buffers, files, directories, and media (images/audio) to the LLM context.
• Integrated Tooling: Built-in commands for code review, commit-message generation, translation, and proofreading.
• Security & Policy: Includes Data Loss Prevention (DLP) scans and filesystem checks to block or warn on sensitive tool actions.
• Multimodal Input: Ability to send images and audio recordings directly to the model.