The core idea is simple: treat the repository itself as the durable working memory for software development, not just as a place to store source code. Fossil is a strong base for this because the repo is already a structured SQLite database with built-in versioning, wiki, chat, tickets, web UI, and artifact history.
This fork is exploring how to extend Fossil into an AI-native development environment with:
- repository-backed provenance for agent actions - automatic micro-commits with prompt/rationale metadata - a tiered knowledge system that keeps raw notes, working context, draft syntheses, and durable atomic concepts - semantic retrieval over repository knowledge using embeddings and vector search - a web-first interface for chat, task flow, knowledge inspection, and change review
The knowledge-management model is based on a pool strategy: ideas bubble up when they are repeatedly retrieved, referenced, or validated, and sink when they cool down. The goal is not just better RAG, but a self-maintaining project memory that stays useful over time instead of turning into an unstructured log of prompts and chats.
The current repo already contains early pieces of this direction:
- AI-specific SQLite schema for context, notes, vectors, and policy - provenance capture tied to commits - local agent chat integration in the Fossil web UI - note storage plus semantic indexing/search plumbing - docs for context assembly, tiers, steering, constitution, metrics, and UI surfaces
The design principles are:
- Fossil remains the source of truth - SQLite is the substrate - minimal dependencies - model/provider agnostic interfaces - structured reasoning and provenance captured as retrievable repository knowledge - strong provenance and inspectability
What I need help with:
- Fossil/C development - SQLite schema and query design - vector search and retrieval quality, including sqlite-vss integration or equivalent approaches - web UI/UX for repository-native agent workflows - knowledge promotion/demotion logic ("bubbling and sinking") - background jobs, automation, and testing - product thinking around what an AI-native SCM/workbench should actually be
If you are interested in version control, SQLite, local-first tools, knowledge systems, or agent tooling with auditable provenance, I would like to talk.
Repository docs are in `doc/ai/` and `doc/specs/` in this repo. If there is interest, I can also write up a more concrete architecture note showing the current implementation, the missing pieces, and the contributor roadmap.