Aiming to provide a seamless and privacy driven chatting experience with open-sourced technologies(Ollama), particularly open sourced LLM's(eg. Llama3.1, Phi-3, Mistral). Focused on ease of use.
LLocal can be installed on Windows, Mac and Linux.
- Llocaly store chats.
- Llocal utilizes Ollama which ensures that from processing to utilizing everything happens on your machine LLocally.
- Seamlessly switch between models.
- Easily pull new models.
- Image upload for models that support vision.
- Web search (i.e Website scraper aswell as duckduckgo search inbuilt) for all models.
- Chat with Files with persistence through vector db's being stored llocally. (Supported file types are PDF, PPTX, DOCX, CSV & TXT)
- Responses are rendered as markdown (Supporting Code Blocks with syntax highlighting, tabular formats and much more).
- Multiple themes (5 themes all suporting both light and dark mode)
- Seamless integration with Ollama, from download to install.
- Chat with images ✅
- Web Search ☑️ (purple because, it still can be improved)
- Retrieval Augmented Generation/RAG (with single PDF's) ✅
- Multiple PDF chat
- Code live preview for web based code (Something like what Claude Provides)
- Text to Speech Models (only if we can get to be similar to a human like response)
- Community wallpapers
- Community themes (something like what spicetify does)
- Lofi Music (this would be optional)
- Speech to text (Do we really need it?)
- Conversations like those with ChatGPT (Speech to text input and text to speech output, but the aim would be low-latency)
- Chat with chats ?! (Not sure)
At some point: would want to pivot LLocal in a different direction... (Although would need to discuss this with the users.)
LLocal's builds are unsigned at the moment, meaning there will be an unknown publisher alert on Windows and Mac. But, on mac it does not open because the code is unsigned and to solve this issue you can do either of the following:
-
Running a manual build by cloning the repo and then running the
npm run build:mac:arm
for m series ornpm run build:mac:intel
for intel based macs. When you build it on your own, that time apple does not throw the error. I know this is inconvenient but the build does take at max a few minutes. -
Incase, you don't want to build it by yourself then you can also try the Universal build that seems to be more stable than the separate builds, but then you'd get the developer is not verified error which can be by passed by following this video .
The link to the mac universal build is this.
LLocal is an Electron application with React and TypeScript.
$ npm install
$ npm run dev
# For windows
$ npm run build:win
# For macOS (m-series)
$ npm run build:mac:arm
# For macOS (intel-chips)
$ npm run build:mac:intel
# For Linux (Supported now!)
$ npm run build:linux
You can refer to the Contribute.md