I've been thinking for a while about running Claude code as a general purpose personal assistant agent that has a) memory about me b) access to my main working tools and c) it's own computer and subagents or sub systems do process things on it's own.
I've had these notes for a while, and have decided to publish them as a forcing mechanism to actually build this, and to crowdsource answers to some of my open design questions. I already built the MVP version of this for my CRM, but want to expand it to include all of the ideas below, and more.
Problems:
There are many tasks that I would like an assistant to help we with but they lack context, especially the precious context contained in my Obsidian vault that's not easy to reach from somewhere else than the computer
I love Claude Code but I can only reach it in my terminal - I would like to be able to access it assistant through Telegram or Whatsapp
There are a bunch of tasks I do often and have my own ways of doing, I want the assistant to remember this and to be able to improve over time at my suggestions
I would like to use voice inputs to create tasks, emails, drafts, calendar events etc. Ramble for a while and that queues up a number of things to execute and triages them accordingly.
Implementations ideas
I want to use Claude Code because it's so powerful, can use MCPs, can manage files and multi-modal input
I need to find a way to get an instance running with its own files in a place where it's always on for me to reach - hardest problem I'm facing for now.
Using sub-agents for specific tasks, but I'm not sure how to make the subagents to improve themselves. An alternative might be to create subfolders for each types of process and and a claude.md file detailing processes and how to approach in each folder
I want to make sure I'm not making this too complex to implement, and that I start with simple use cases and gradually improve in complexity.
Example tasks
I've been thinking for a while about running Claude code as a general purpose personal assistant agent that has a) memory about me b) access to my main working tools and c) it's own computer and subagents or sub systems do process things on it's own.
I've had these notes for a while, and have decided to publish them as a forcing mechanism to actually build this, and to crowdsource answers to some of my open design questions. I already built the MVP version of this for my CRM, but want to expand it to include all of the ideas below, and more.
Problems:
There are many tasks that I would like an assistant to help we with but they lack context, especially the precious context contained in my Obsidian vault that's not easy to reach from somewhere else than the computer
I love Claude Code but I can only reach it in my terminal - I would like to be able to access it assistant through Telegram or Whatsapp
There are a bunch of tasks I do often and have my own ways of doing, I want the assistant to remember this and to be able to improve over time at my suggestions
I would like to use voice inputs to create tasks, emails, drafts, calendar events etc. Ramble for a while and that queues up a number of things to execute and triages them accordingly.
Implementations ideas
I want to use Claude Code because it's so powerful, can use MCPs, can manage files and multi-modal input
I need to find a way to get an instance running with its own files in a place where it's always on for me to reach - hardest problem I'm facing for now.
Using sub-agents for specific tasks, but I'm not sure how to make the subagents to improve themselves. An alternative might be to create subfolders for each types of process and and a claude.md file detailing processes and how to approach in each folder
I want to make sure I'm not making this too complex to implement, and that I start with simple use cases and gradually improve in complexity.
Example tasks
Put all my invoices in a folder with the right title format based on the content of the attachement
Answer questions based on content from my obsidian - "when did I speak to xxx for the last time and what did we discuss"?
Please create a calendar event for September 13th called meeting with Lewis and invite Lewis - the event needs the right person + a meeting link
Please write a message to my fiancée telling her I'm on my way and will arrive at 8pm - needs to be in my tone and in french, as inferred from the context
Create a new obsidian note titled 'Observations from my trip in Asia' with the following ideas in there - idea 1; idea 2, etc
Put all my invoices in a folder with the right title format based on the content of the attachement
Answer questions based on content from my obsidian - "when did I speak to xxx for the last time and what did we discuss"?
Please create a calendar event for September 13th called meeting with Lewis and invite Lewis - the event needs the right person + a meeting link
Please write a message to my fiancée telling her I'm on my way and will arrive at 8pm - needs to be in my tone and in french, as inferred from the context
Create a new obsidian note titled 'Observations from my trip in Asia' with the following ideas in there - idea 1; idea 2, etc
Tools
Obsidian
Todoist
Gmail
Telegram
iMessage
Notion
Slack
Tools
Obsidian
Todoist
Gmail
Telegram
iMessage
Notion
Slack
Monetizing open social graphs
Traditional social networks offer creators the possibility to monetize mainly one-to-many relationships.One-to-many monetizationThe basic description...
Monetizing open social graphs
Traditional social networks offer creators the possibility to monetize mainly one-to-many relationships.One-to-many monetizationThe basic description...
daily (mostly) publication. I write to learn, sharing ideas and notes for what I find interesting
daily (mostly) publication. I write to learn, sharing ideas and notes for what I find interesting
Subscribe to one small idea
Subscribe to one small idea
Share Dialog
Share Dialog
>100 subscribers
>100 subscribers
If you have a computer always running, you can run Claude code from your mobile device using Happy Coder. @xr0am- on Farcaster also figured out how to run it in a VPS so it didn’t have a computer at home that was always running
Exploring the vision for a personal assistant powered by Claude code, @luc shares ideas on memory integration, accessibility through messaging apps, and the ability to learn from user habits. The post details challenges such as providing context from documents and utilizing multi-modal inputs while suggesting implementation options like sub-agents and self-improving tasks. Come along on this journey to design an efficient assistant that simplifies daily tasks and enhances productivity!
If you have a computer always running, you can run Claude code from your mobile device using Happy Coder. @xr0am- on Farcaster also figured out how to run it in a VPS so it didn’t have a computer at home that was always running
Exploring the vision for a personal assistant powered by Claude code, @luc shares ideas on memory integration, accessibility through messaging apps, and the ability to learn from user habits. The post details challenges such as providing context from documents and utilizing multi-modal inputs while suggesting implementation options like sub-agents and self-improving tasks. Come along on this journey to design an efficient assistant that simplifies daily tasks and enhances productivity!