Inspiration
There was a twitter post mentioning how nobody has built a widget app with the little FaceID emoticon. 😊
What it does
Dashi aims to solve the shortcomings of Siri using modern AI functionalities. Using features like tool calling and MCP servers, we give an AI assistant like Siri much more capability.
How we built it
We built the agent powering Dashi with Letta's new sleep time voice agent.
Challenges we ran into
The mix of environments (cloud apis, local functions) caused a lot of trouble during Dashi's development. Letta's integration with other services such as Vapi also had some issues, forcing us to debug for hours.
Accomplishments that we're proud of
We built a clean, opinionated UI that the user can easily navigate to use their assistant. Dashi is lightweight (thanks to Tauri) and lives on your system idly until you need it.
What we learned
MCP is not completely fleshed out and the agent ecosystem has a far way to go in standardizing how agents are built. Cut and paste integrations are finnicky and hard to debug without a constant support team (Thank you Charles and Sarah!)
What's next for Dashi
Make it modular by allowing the user to customize their own assistant (add their own tools, bring their own MCP servers, etc.) Improve the UI/UX across all of Dashi's capabilities, ensuring a seamless experience for the user.
Built With
- composio
- exa
- figma
- framer-motion
- groq
- letta
- lottie
- python
- rust
- typescript
- whisper
Log in or sign up for Devpost to join the conversation.