Claude Code feels amazing in codebases. Knowledge bases should feel the same.
Palo Alto, CA
Joined January 2025
- Tweets665
- Following504
- Followers321
- Likes423
Pinned Tweet
Claude Code changed how I work with codebases.
I want the same experience for personal knowledge:
AI-native writing and retrieval.
We can connect Notion or Obsidian to Claude today.
But honestly, most note apps don’t really want to become an AI retrieval layer.
My husband builds vector databases, so maybe we’re unusually comfortable with that lol😂.
So we decided to build StashBase:
🧠 a local knowledge base with built-in Claude Code
⚡ continuously indexed
🔌 MCP-compatible retrieval
Still very early (day 3), but excited to explore what an AI-native knowledge base should actually feel like.
github.com/liliu-z/stashbase
StashBase retweeted
Shipped an upgrade to the @stashbaseai Agent panel.
- cleaner tool activity
- visible approval prompts
- one-click file outputs
- better chat history replay.
check this out !! any suggestions on making this better is much appreciated !!
Cmd+F is string matching.
The search in your head is "that thing about cache invalidation I read last month."
The gap between those two is an index.
Text files should've had one a decade ago.
Add up all the time you've spent naming files. It's a book.
"final_v2", "Untitled 3", "notes copy" — every one of those is future-you failing to find it.
Let an index remember the content. Stop making filenames do a job they were never good at.
Folders are 1970s UX debt.
Every time you drop a note into some directory, you're betting on which folder future-you will look in.
You lose that bet every time.
Text shouldn't be organized by path. It should be organized by an index.
A few things I learned from Reddit users while discussing local file search for AI agents:
1. The real gap in local AI file search isn’t file access. It’s the lack of a reusable knowledge index layer.
2. Continuous indexing is what turns a demo into a real workflow. Local knowledge needs to stay useful over time.
3. StashBase shouldn’t just be “PDF chat.” It should be a local context backend for AI agents.
4. One-time indexing is easy. The real value is making files still searchable after they are added, edited, moved, or deleted.
How many times have you explained the same project to ChatGPT or Claude this week?
More than 10?
Or is it just me?
I'd love to connect with more CS students.
If you're studying CS and building side projects, drop what you're working on below 👇
I'd love to check it out.
I'm building a local AI search tool.
What's the most cursed document folder on your computer?
Scanned PDFs, OCR nightmares, weird layouts, docs AI usually struggles with...
Drop the worst one in the comments. I’ll pick a few to test.
Built this for our own team.
We needed real-time Chinese → English speech translation, so I hacked together a tiny macOS app.
🎙️ Chinese speech in
🤖 Gemini Live translates
🔊 English audio out → Zoom / Google Meet / Discord / Feishu
I wasn't planning to release it.
But if other bilingual teams have the same problem—or need support for different languages—I’d be happy to turn it into something more polished.
If the idea sounds interesting to you, let's connect.
GitHub: github.com/0-bingwu-0/live-i…
One thing I've found surprisingly helpful during my startup journey:
Whenever I feel stuck or anxious, I talk to ChatGPT.
A small trick that changed those conversations was giving it Stanford CS183 as context.
Instead of generic startup advice, it started reasoning more like an experienced founder. The conversations became much more grounded, and honestly, it helped ease a lot of my anxiety.
So I converted the entire course into clean Markdown and open-sourced it.
If you'd like a copy, reply CS183 and I'll send it over.
Strong AI/RAG/MCP background!
About me:
I'm a final-year Student
My experience includes AI Agents, MCP (Model Context Protocol), RAG Systems, LLM Orchestration, FastAPI, React, Next.js, TypeScript, Python, Docker, PostgreSQL, Vector Databases, and modern AI workflows.
I've worked on AI agent frameworks, autonomous workflows, document extraction systems, and real-time AI applications. I actively search for opportunities every day, and while doing so, I share job openings with the community to help others as well.
Currently looking for remote Internship
If you know of any openings or can provide a referral, I'd be grateful. 🙏
Love this. We’re also looking to connect with CS students interested in AI-native tools.
We’re building StashBase, an open-source AI memory project.Happy to chat with anyone who wants to learn by building.
I am looking to expand my IP and I need the right people to help me build it bigger.
I am Hiring:
- Video Editor
- Project Managers
- Researchers
- Social Media Managers
- Content Writers
- Community moderators
Drop a comment with what you do.
And if you think you can add value beyond what I have listed, drop that in the comment too.
StashBase retweeted
🚨Hiring 1–2 AI Fellows! 🚨
We’re encoding messy institutional knowledge into a knowledge graph for private capital markets.
Our agentic system sits on top of this graph, so the work spans scalable extraction, retrieval, entity linking, provenance, and verification.
DMs open.
Hey @X 👋
Excited to #connect with people who are passionate about technology and building cool things.
I’m currently exploring:
• SaaS
• Data Science
• Startups
• Building in public
• AI Tools
• Tech Communities
Looking forward to learning, building and #connecting with amazing people here✨
Agents operating on Markdown files don't prove databases are unnecessary.
They may prove that most databases were optimized for the wrong workload.
Experimenting with a 📢 public build map 🗺️ for my local-first knowledge base built around Claude Code workflows.
As implementation gets easier with agents, project context and structure start to matter a lot more.
So I started making more of the project structure public while building:
architecture, workflows, planned modules, retrieval flow, MCP integration, system boundaries, etc.
Would love feedback — and also curious how other people are organizing AI-assisted projects/workflows.
stashbase.ai/build-map/
I kept trying to turn Obsidian into an AI workspace with HTML, terminal, and retrieval plugins.
Eventually I realized the issue wasn’t missing plugins. The assumptions were different.
Obsidian is fundamentally designed around humans manually writing and organizing notes.
After Claude Code, manually writing and managing everything myself just feels hard to return to.