Vibe-Kit
Vibe-Kit listens to your conversations, generates notes called "memories" and let's you choose an "action" that triggers an MCP server to "vibe code" turning your memories into open source Githubs.
Created on 16th March 2025
•
Vibe-Kit
Vibe-Kit listens to your conversations, generates notes called "memories" and let's you choose an "action" that triggers an MCP server to "vibe code" turning your memories into open source Githubs.
The problem Vibe-Kit solves
You generate great ideas all the time, but many ideas are lost because you don't take notes & and follow up with real action. You fail 100% of the projects that you don't start. With Vibe-Kit, every idea is captured as a "memory" and with just one click of an "action" it is converted into a detailed project on GitHub and the project development plan is kicked and connecting via MCP to all of the necessary resources. Many more actions are possible - not just vibe coding - and we testing ordering from UberEats via API. If you're an ideas person, you probably thought of a great idea before it took off, but you didn't take action. If you have trouble getting started on new projects, the vibe kit will get you started and jumpstart development.
User Interaction and Data Flow
Download the OMI app on your phone connected to your Bluetooth necklace or press use phone mic. The vibe kit app is available on the OMI App Store and you can immediately start sending data to your vibe server to trigger new GitHub projects to be built on the fly while you have conversations. Just install the app and start talking. Users can also log into the Vibe Kit website and see how many memories they'd captured and click "actions" for each memory they create.
The project architecture and development process
The Vibe-Kit connects to the Webhook API provided by the open source Omi device. The Omi is a small necklace wearable microphone that connects to your phone via Bluetooth using an open source mobile app the audio recording is streamed live to a transcription service Deepgram and the resulting transcription is sent back to the device to the phone as well as sent to be stored in a vector database known as pinecone. Omi sends the transcript to the Vibe-Kit server where the memories are stored in a database. Users can then choose to burn the memory or store the memory as well as send the memory to their new contacts. The action item app takes those memories and breaks them down into different projects with different action items.
We use farcaster for posting memories to social. We use Claude for AI inference and creating project issues. We use pinecone for a vector database. We also use open AI in some cases to generate text to post on Farcaster. We use postgres for our database on our server. We're using next JS as our primary job care framework.
Product Integrations
We used the Coinbase AgentKit to interact with the blockchain. Users can like a project or fund a project on chain.
We used the UberEats API to place orders for food and record those transactions on chain.
Key differentiators and uniqueness of the project
Vibe-Kit does the DevRel job by generating sample code insanely fast and quietly in the background while the DevRel "vibes" with hackers at a hackathon. Vibe-kit can do much more now, with customizable actions linked to MCP servers - for example ordering from UberEats.
Our project is the first of it's kind to use real live conversation data to "vibe code" a project, and we designed it to automate the job of Developer Relations Engineers.
Trade-offs and shortcuts while building
We are still using centralized cloud serviers for transcription, inference, and vector database for speed of development. Vitalik says “this is why we need edge compute.”
We consider these centralized cloud services to be reasonable trade-offs in order to use the open source Omi project. We plan to leverage xtrace.ai for a private vector database, Mizu for edge data, and NEAR AI for a more privacy focused LLM.
Additional Features
The concept was created at ETH Denver, but we ran out of time trying to get all the pieces connected. We did a lot of new work this week at ETHSF, adding "actions" for each memory, adding a Community Leaderboard, adding a Bounty Board, building our own MCP server, getting the front end working and reliable. We integrated the Uber Eats API to place orders on chain via the Agent Kit. We connected to the Base Mainnet instead of Hedera and Flow.
Tracks Applied (2)
Build a new and innovative consumer experience on Base
Base
Most Innovative Use of AgentKit
Coinbase Developer Platform