Please note this repo will migrate to https://www.piwheels.org/project/swamix/ as internal toolkit. why?
This Copilot is highly customizable multi purpose with agentic capabilities, which assists user based on their personalized queries primarily with verbal query, jarvis style.
It can be used for a variety of tasks such as
- Note Taker/Maker, with speech to text
- meme explainer, OCR image to text
- document analysis, presentations exporter
- realtime online conference support,
- content writing, translation, quick web research
to keep this list small, what you can do with a mouth, eyes, hands and secondary brain, you can do with this copilot.
Currently a hackathon proof of concept, but can be extended to a full fledged product. please open issue and tell us what you do, we will add functionality carefully so it may benefit people who do same task as you!
Features a plug-n-play architecture with many batteries included so a user can easily add or remove features as per their needs (you can keep as light or as heavy).
- BYOM : Bring your own model / finetune
- BYOT : Bring your own tools (Agent)
- BYOC : Bring your own context / data
- BYOP : Bring your own prompt / template
first clone repo to a folder:
git clone https://github.com/nikhil-swamix/sambanova-copilot.git
cd sambanova-copilot
pip install -r requirements.txt
python copilot.py
Try to directly run the run__.ps1 or .sh. here _ is platform. it will install dependencies and run the app. run 2 times if it fails the first time, (troubleshoot tip).
download fow win Go to releases section and download the executable, and double click!
# Cd to directory after git clone
git clone https://github.com/nikhil-swamix/sambanova-copilot.git
cd sambanova-copilot
poetry install
poetry run python copilot.py
A great care has been taken to make it distributable and easy to use, please open issue in repo if any encountered.
make sure you have SAMBANOVA_API_KEY, GROQ_API_KEY , configured in .env , some keys are provided.
- Press 1 (🎤) - Start/stop voice recording
- Press 2 (📸) - Capture current screen
- Press 3 (🤖) - Send to AI for processing, it will show response in dedicated popup
- Press 4 (⌨️) - Auto-type response (100 chars/sec)
- Toggle Clipboard - Use copied text as additional context
- Toggle Workspace - Reference local documents
- Toggle Web Search - Include internet results
- Reset (🔄) - Clear current session data
- Code Review: Screenshot code + voice ask "explain this code"
- Meeting Notes: Record conversations for instant transcription and summary
- Meme Explainer: Capture meme image + ask for cultural context
- Document Analysis: Screenshot text + ask for key points summary
- Technical Support: Capture error screen + ask for solutions
- Content/Blog Writing: Dictate rough ideas for AI enhancement
- Translation Helper: Record foreign language for translation and tips
- Quick Research: Combine clipboard articles + voice questions
- UI/UX Review: Screenshot designs for instant feedback
- Email Assistant (with plugin): Dictate main points for professional email drafts
Pro Tip: Combine voice + image + clipboard for best results
- generate executive reports after scanning few webpages, Save Generated content to PDF/Markdown (artifacts)
- Does not require API for web search, use local browser for free search! (msedge, chrome, firefox) !
- Can use local documents as context for better results
- Can use clipboard text as additional context
- Can use voice recording as primary instruction.
- Automatic Tool Decision Based on user query (Agentic/Smart Mode) (Experimental, please comment out code if any issue)
Configuration help
KEYS_WILL_BE_DELETED_POST_HACKATHON_DATE recommended to get your own keys, these keys are in .env file
https://cloud.sambanova.ai/apis
Used for text to speech
https://console.groq.com/keys
https://dash.voyageai.com/api-keys
- Reasoning mode (some challenge with llama to act as thinker, in progress)
- Save history and submit to SambaStudio finetune job so model just "gets it"
- ui based prompt editor , add remove rules in system prompt on the fly
- Share json backed OPEN API compatible Workflows using "SambaHub" maybe.
- Please open issues if any encountered. will be added and try to resolve.