# Weft
> Turning ChatGPT conversations into a searchable, structured second brain.
## What is this?
Weft is an open-source system for reconstructing, organizing, and understanding exported ChatGPT conversations.
The goal is not just to store chats.
The goal is to transform thousands of scattered AI conversations into:
- a knowledge graph
- a research archive
- a personal context engine
- a second brain
- and eventually an AI memory system
Instead of conversations dying inside chat windows, Weft tries to make them:
- searchable
- interconnected
- reusable
- analyzable
- and useful for long-term thinking.
---
# Why I’m Building This
Every serious AI user eventually hits the same problem:
- thousands of conversations
- buried ideas
- forgotten insights
- repeated questions
- no structure
- no continuity
ChatGPT remembers almost nothing across time.
Humans don’t either.
So the idea is:
> What if conversations could become persistent knowledge?
Not bookmarks.
Not notes.
Actual evolving context.
---
# Vision
Weft aims to become a system that can:
- Parse exported ChatGPT conversations
- Reconstruct conversation trees
- Extract ideas, topics, projects, and entities
- Build semantic search over conversations
- Connect related thoughts automatically
- Generate knowledge graphs
- Create long-term AI memory
- Integrate with Obsidian and local knowledge systems
- Enable AI-assisted reflection and research
Eventually:
- AI agents that understand your history
- Context-aware assistants
- Personalized reasoning systems
- Self-organizing research archives
---
# Core Idea
Chat logs are not just messages.
They are:
- thinking traces
- learning history
- project evolution
- research logs
- decision timelines
- idea graphs
Weft treats conversations like structured knowledge instead of disposable text.
---
# Current Features
## Parsing ChatGPT Export Data
- Reads exported `conversations-000.json`
- Reconstructs branching conversations
- Preserves message hierarchy and metadata
## Obsidian Integration
- Converts chats into markdown
- Builds vault-ready structures
- Enables backlinking and note navigation
## Conversation Reconstruction
- Restores message flow from node mappings
- Creates readable timelines
- Preserves assistant/user roles
## Local Knowledge Storage
- Fully local-first workflow
- Your data stays yours
- No dependency on cloud vector DBs
---
# Planned Features
## Semantic Search
Find ideas instead of keywords.
## Embedding Pipelines
Vectorize conversations for contextual retrieval.
## Knowledge Graphs
Automatically connect:
- projects
- concepts
- people
- research topics
- recurring patterns
## AI Context Engine
Provide historical context to local LLMs and agents.
## Memory Compression
Summarize years of conversations into reusable knowledge.
## Multi-Source Integration
Future support for:
- Claude exports
- Gemini chats
- emails
- notes
- PDFs
- research papers
---
# Tech Stack
## Backend
- Python 3.11
- FastAPI
## Data Processing
- Pandas
- Pydantic
- NetworkX
## NLP / AI
- Sentence Transformers
- FAISS / ChromaDB
- LangChain (possibly)
- local embeddings
## Frontend (planned)
- React
- TypeScript
## Knowledge Layer
- Obsidian
- Markdown
- Graph-based linking
---
# Philosophy
Weft is built around a few ideas:
### AI conversations are valuable data
Most people waste them.
### Knowledge should compound
Ideas should connect over time.
### AI should enhance thinking
Not replace it.
### Local-first matters
Your thoughts should belong to you.
### Context is everything
Intelligence without memory is shallow.
---
# Long-Term Goal
The long-term goal is to build something between:
- a second brain
- a research operating system
- and a persistent AI memory layer
A system where:
- your ideas evolve over years
- AI understands your projects
- context accumulates instead of disappearing
---
# Current Status
Early-stage experimental project.
Right now the focus is:
1. understanding ChatGPT export structure
2. reconstructing conversations correctly
3. building clean markdown pipelines
4. designing scalable architecture
5. preparing for semantic retrieval and AI memory systems
---
# Example Workflow
```bash
Export ChatGPT data
↓
Parse conversation JSON files
↓
Reconstruct conversation trees
↓
Convert to markdown
↓
Store in Obsidian
↓
Generate embeddings
↓
Semantic search + knowledge graph
↓
AI memory/context engine- Build something genuinely useful
- Learn deeply instead of tutorial-copying
- Understand AI memory systems
- Explore knowledge engineering
- Create infrastructure for future AI agents
- Open-source the entire process
Inspired by:
- second brain systems
- knowledge graphs
- semantic search
- AI memory architectures
- personal knowledge management
- research workflows
- long-term thinking systems
Still heavily experimental.
But if you care about:
- AI memory
- knowledge systems
- semantic retrieval
- local-first AI
- personal context engines
- Obsidian workflows
feel free to explore, fork, or contribute.
Most AI conversations disappear.
Weft exists because maybe they shouldn’t.