Ever found yourself wanting to add a new AI provider to ollama-buddy? (probably not I would guess 🙂), only to realise you'd need to write an entire Elisp module? Or perhaps you're running a local inference server that speaks the OpenAI API, but can't be bothered with the ceremony of creating a dedicated provider file?

There is now an in-buffer replace feature in ollama-buddy, so now an ollama response can work directly on your text, streaming the replacement in real-time, and giving you a simple accept/reject choice!, I have also added an smerge diff inline if desired to show the differences and give the user the ability to accept or reject

One of the fundamental limitations of local LLMs is their knowledge cutoff - they don't know about anything that happened after their training data ended. The new web search integration in ollama-buddy solves this by fetching current information from the web and injecting it into your conversation context. Ollama has a specific API web search section, so it has now been activated!

One of the things that has always slightly bothered me about chatting with a local LLM is that it only knows what it was trained on (although I suppose most LLMs are like that) . Ask it about your own codebase, your org notes, your project docs - and it's just guessing. Well, not anymore! Ollama Buddy now ships with proper Retrieval Augmented Generation support built-in

Tool calling has landed in ollama-buddy!, it's originally not something I really thought I would end up doing, but as ollama has provided tool enabled models and an API for this feature then I felt obliged to add it. So now LLMs through ollama can now actually do things inside Emacs rather than just talk about them, my original "do things only in the chat buffer and copy and paste" might have gone right out the window in an effort to fully support the ollama API!

I've had a little free time recently (figuring out this baby stuff!) and thought I would spend time revisiting and refining my AI assistant ollama-buddy I've been playing around with agentic coding and keeping up-to-date on the rapid development of the Emacs AI package landscape and I think I have refined in my own mind my idea of what I would like to see in an Emacs AI assistant. The headline change regarding the latest release of ollama-buddy is GitHub Copilot integration; the rest of the work is about smoothing the UI and simplifying day-to-day use.

Having another look at my AI assistant - ollama-buddy, its been a while and it seems ollama has moved on since I started creating this package last year, so I have developed a new roadmap and the first step is to add ollama cloud models!

Ollama Buddy now supports Mistral's Codestral - a powerful code-generation model from Mistral AI that seamlessly integrates into the ollama-buddy ecosystem.

After months of development and refinement, I'm excited to announce Ollama Buddy v1.0 - an Emacs package that simply interfaces mainly to ollama, for local LLM usage, but can be integrated to the major online players. This project initially started as a simple integration with Ollama and since then has somewhat evolved into a more fully fully-featured AI Emacs assistant. The main focus with this package is a front facing simplicity but hiding (hopefully) all the features you would expect from an AI chatbot - wait I hate that term, I mean, assistant :). There is also the ability to craft a customizable menu system for different roles.

I had a blast developing this package and next up is RAG!. I saw recently that a package called vecdb was introduced into the package ecosystem to help with the storage of vector embeddings, so as ollama can return embedding vectors for semantic search I thought I would combine my package, vecdb, also probably initially a PostgreSQL database with a pgvector extension and ollama into something that could ingest files directly from Emacs. I think I have figured this out now, I just need to do it (when the baby is asleep, probably!)

There have been quite a few updates recently. The main highlights include support for attachments, so you can push a file to the chat directly from dired for potential inclusion in your next query. Vision support has been added for models that can handle it. If you supply the path to an image file in the chat, it will be processed. This means you can now, for example, extract text from images using models like o:gemma3:4b. I've also introduced the ability to save user system prompts. If you have a favorite prompt, or have crafted one that works especially well for you, you can now save it by category and title in a simple Org format for later recall. Prompt recall now works the same way as Fabric patterns and Awesome ChatGPT prompts. This makes it much easier to display the currently used system prompt concisely in the status bar, as it will be based on the prompt title (and thus likely the role). What else? Oh yes, I received a request for better context tracking. Now, when context is nearing full capacity, or has exceeded it, it will be indicated in the status bar! That’s probably it for the major changes. There was also some refactoring, but you probably don't care about that. Anyway, here is the full list of changes:

More improvements to ollama-buddy https://github.com/captainflasmr/ollama-buddy

Several improvements in the latest Ollama Buddy updates (versions 0.9.21 through 0.9.35):

• Added ollama-buddy-awesome.el to integrate Awesome ChatGPT Prompts.

ollama-buddy-awesome is an ollama-buddy extension that integrates the popular Awesome ChatGPT Prompts repository, allowing you to leverage hundreds of curated prompts for various tasks and roles right within your Emacs environment, I thought that since I have integrated the fabric set of curated prompts, so then why not these!

This week in ollama-buddy updates, I have mostly been experimenting with ChatGPT integration! Yes, it is not a local LLM, so not ollama, hence entirely subverting the whole notion and fundamental principles of this package! This I know, and I don't care; I'm having fun. I use ChatGPT and would rather use it in Emacs through the now-familiar ollama-buddy framework, so why not? I'm also working on Claude integration too.

This week in ollama-buddy updates, I have been continuing on the busy bee side of things. The headlines are :

• Transient menu - yes, I know I said I would never do it, but, well I did and as it turns out I kinda quite like it and works especially well when setting parameters.
• Support for fabric prompts presets - mainly as I thought generally user curated prompts was a pretty cool idea, and now I have system prompts implemented it seemed like a perfect fit. All I needed to do was to pull the patterns directory and then parse accordingly, of course Emacs is good at this.
• GGUF import - I don't always pull from ollama's command line, sometimes I download a GGUF file, it is a bit of a process to import to ollama, create a model file, run a command, e.t.c, but now you can import from within dired!
• More support for the ollama API - includes model management, so pulling, stopping, deleting and more!
• Conversation history editing - as I store the history in a hash table, I can easily just display an alist, and editing can leverage the sexp usual keybindings and then load back in to the variable.
• Parameter profiles - When implementing the transient menu I thought it might be fun to try parameter profiles where a set of parameters can be applied in a block for each preset.

More improvements to ollama-buddy https://github.com/captainflasmr/ollama-buddy The main addition is that of system prompts, which allows setting the general tone and guidance of the overall chat. Currently the system prompt can be set at any time and turned on and off but I think to enhance my model/command for each menu item concept, I could also add a :system property to the menu alist definition to allow even tighter control of a menu action to prompt response.

Continuing the development of my local ollama LLM client called ollama-buddy… https://github.com/captainflasmr/ollama-buddy The basic functionality, I think, is now there (and now literally zero configuration required). If a default model isn't set I just pick the first one, so LLM chat can take place immediately.

Some improvements to my ollama LLM package… With the new multishot mode, you can now send a prompt to multiple models in sequence, and compare their responses, the results are also available in named registers.