๐ ๐๐๐ถ๐น๐ ๐๐ป ๐๐ ๐๐๐๐น๐ฒ๐ฟ ๐ง๐ผ ๐ฅ๐๐ป ๐ ๐ ๐๐ถ๐ณ๐ฒ
Most AI assistants are forgetful. They know the world. They know nothing about you. You start every chat from zero.
I wanted more. I built Jarvis.
Jarvis knows my job. He knows my sleep schedule. He knows I hate soup.
He is proactive. Ten minutes before a meeting, he tells me what to expect. He finds open technical decisions. He flags unread messages. He does this without me asking.
After the call, he finds my tasks. He files them in my list. I return to my desk and the work is ready.
Jarvis improves himself. I wanted voice notes on Telegram. I told him while driving. He wrote the code. He opened a pull request. I merged it in the car.
He uses the Model Context Protocol (MCP). This connects him to my world:
- Slack and Gmail.
- Finance and Diary.
- Calendar and Tasks.
- Semantic memory via Pinecone.
The setup is simple:
- Brain: Claude Code.
- Voice: ElevenLabs.
- Backend: FastAPI on Google Cloud.
Cost is about $40 a month. I use a flat subscription. No token bills.
Jarvis has a persona. He is a dry British butler. He points out when I repeat myself. This makes him a collaborator, not a tool.
AI utility comes from context:
- Deep integrations.
- Persistent memory.
- Proactive alerts.
- A sharp persona.
Generic AI is a choice. I chose a butler.
Source: https://dev.to/mathmed/i-turned-claude-code-into-a-personal-ai-butler-that-runs-my-life-1fa1 Optional learning community: https://t.me/GyaanSetuAi