Version 1.14.2 of package Ellama has just been released in GNU ELPA. You can now find it in M-x list-packages RET.
Ellama describes itself as: ============================== Tool for interacting with LLMs ============================== More at https://elpa.gnu.org/packages/ellama.html ## Summary: [file:https://img.shields.io/badge/license-GPL_3-green.svg] [file:https://melpa.org/packages/ellama-badge.svg] [file:https://stable.melpa.org/packages/ellama-badge.svg] [file:https://elpa.gnu.org/packages/ellama.svg] Ellama is a tool for interacting with large language models from Emacs. It allows you to ask questions and receive responses from the LLMs. Ellama can perform various tasks such as translation, code review, summarization, enhancing grammar/spelling or wording and more through the Emacs interface. Ellama natively supports streaming output, making it effortless to use with your preferred text editor. The name "ellama" is derived from "Emacs Large LAnguage Model Assistant". Previous sentence was written by Ellama itself. <file:imgs/reasoning-models.gif> ## Recent NEWS: 1 Version 1.14.2 ════════════════ • Add project-local skill pointers for changelog and commit-message workflows. This makes repository-specific Codex skills discoverable from `.codex/skills'. • Add a `commit-message' skill for concise commit message generation from diffs. • Update agent instructions to document the project commit workflow. The workflow now covers pre-commit checks, branch creation from `main', iterative check/fix/commit/push cycles, documentation updates, changelog generation, version bumps, and when to ask for clarification. 2 Version 1.14.1 ════════════════ • Refactor transient model handling and provider construction. Generalize transient model handling to support multiple provider types beyond Ollama, including OpenAI-compatible providers. • Implement `ellama-transient-set-model' and `ellama-transient-set-url' for broader provider support. • Add `ellama-transient-reset-model-fields' to reset model-specific parameters. • Refactor `ellama-construct-provider-from-transient' to dynamically handle different provider types. • Improve transient descriptions to dynamically show if values are set to defaults. • Add logic to filter transient menu items based on the selected provider type. • Add support for loading models and filling settings from existing provider defaults and current sessions. • Update tests to cover various provider types and edge cases. • Refactor provider parameter extraction. Introduced `ellama-transient--provider-context-length' to handle context length retrieval. • Removed hardcoded default values for temperature and context length, ensuring parameters are directly retrieved from the provider. • Update documentation. 3 Version 1.14.0 ════════════════ • Implement automatic session context compaction. This feature automatically summarizes chat history when it reaches a certain token threshold to manage context window limits. It includes customization variables for threshold, target size, provider, and turn retention. • Refactor session compaction to be asynchronous using `llm-chat-async`. This introduces helper functions to manage token counts, apply compaction results, and handle asynchronous errors and state changes. • Add a compaction minor mode (`ellama-compaction-mode`) for visual feedback in the mode line when session compaction is in progress. • Refactor session compaction to use synthetic interactions for summaries. Instead of appending the summary to the prompt context, it is now represented as a synthetic assistant interaction. This ensures the original system context is preserved and subsequent compactions correctly replace the previous summary. • Improve session compaction system interaction handling. Added functions to identify, drop, and extract system context from leading system interactions, ensuring they are correctly preserved during the compaction process. • Add session compaction options to the transient menu, including "Compact Current Session" and "Compact Session" commands. • Fix docstring warnings and reformat code. 4 Version 1.13.0 ════════════════ • Implement DLP (Data Loss Prevention) layer. • Add prompt‑injection detection and configurable output‑warning handling. • Add SRT sandbox integration to ellama-tools. • Remove apply_patch tool. 5 Version 1.12.18 ═════════════════ • Add refill-readme target. Updated AGENTS.md to include a new “Refill readme” step in the build instructions. Modified Makefile to add the target to the `.PHONY` list and to define the `refill-readme` rule, which runs Emacs in batch mode to format the `README.org` file. • Refactor org filling for README and NEWS. Introduce a reusable `refill-org` macro that uses `org-element` to fill paragraphs to 80 columns, replacing the previous `fill-region` approach. Update `refill-news` and `refill-readme` targets to use this macro, simplifying the Makefile and ensuring consistent formatting across org files. • Update readme formatting. Minor formatting tweaks. 6 Version 1.12.17 ═════════════════ • Improved new session handling in `ellama-chat' by refactoring session … …
