Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Build your own LLM-native WIKI (knowledge library). Search, extract, summarize, Q&A with contextual RAG, layered knowledge graph, and reinforced memory. Importantly use selected context to automatically generate skills, empowered by Claude subagents + CodeAct pipeline and gated by human review. **Try Live Demo**: https://byo-wiki-demo.vercel.app
| Date | Stars |
|---|---|
| 2026-07-31 | 266 |
| 2026-08-04 | 270 |
| 2026-08-06 | 270 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Build Your Own WIKI
<img width="1672" height="941" alt="ChatGPT Image Jun 22, 2026, 01_55_25 AM" src="https://github.com/user-attachments/assets/5e5243b5-9614-4f5d-9cc2-b54b492a8fe8" />
Turn the web, your files, and loose notes into a **personal wiki** that an LLM agent
keeps coherent: search and summarize pages, ingest them into a contextual vector
index for grounded, cited Q&A, and grow a layered knowledge graph that de-duplicates
entities, builds topics, and writes synthesis pages. One Flask app — web UI, JSON API,
and a `runner.py` CLI. Inspired by LangChain's *llm-wiki* deep-agents example, but it
builds a private, on-disk library instead of syncing to a hub. **Local-first**: every
store is plain JSON/SQLite under `data/`; cloud is opt-in.
[](https://byo-wiki-demo.vercel.app)
## Features
- **Agentic knowledge graph** — a [`deepagents`](https://pypi.org/project/deepagents/)
agent (local filesystem backend, no cloud sandbox) saves passages, extracts entities
and typed relations, canonicalizes duplicates, nests topics, and writes synthesis pages.
- **Contextual RAG** — Anthropic-style *contextual retrieval* over a two-layer HNSW index
(section summaries + chunks), with an LLM re-ranker (precision) or document-aware MMR
(multi-doc recall). Answers are grounded and cited.
- **Memory** — a cross-session store recalled *before* every answer and written back
*after* (observations, 👍/👎, corrections that supersede stale notes); it improves from
use, not just ingestion.
- **Agent skills** — turn selected context into a reusable, *evaluated* skill via a
sub-agent pipeline (understand → analyze → author → eval → gate → refine). Authored by
the latest Claude in-process or via the **Claude Code CLI as a subprocess**, scored by a
deterministic + rubric panel, and **gated behind human review** before it joins the
library — optionally as a durable **LangGraph** build that pauses at the gate and resumes later.
- **MCP, both directions** — connect agents to external MCP servers (Supabase, GitHub, …)
and run BYO-WIKI *as* an MCP server. Reads join the agent; **writes are deny-by-default**.
## Architecture
```mermaid
flowchart TB
subgraph IF["Interfaces"]
direction LR
UI["Web UI · Flask"]
API["JSON API · /api"]
CLI["CLI · runner.py"]
end
ING["Ingestion pipeline<br/>fetch · clean · contextual summaries<br/>chunk · embed · entity / relation / topic extraction"]
subgraph STORE["Stores · local-first under data/"]
direction LR
KG[("Knowledge graph · layers 0–5<br/>source · section · chunk<br/>entity · topic · synthesis")]
VEC[("HNSW index<br/>section summaries + chunks")]
MEM[("Memory · layer 6")]
SKL[("Skill library · layer 7")]
end
subgraph REASON["Reasoning"]
direction LR
RAG["Contextual RAG<br/>retrieve → re-rank / MMR → cite"]
AGENT["Curating agent · deepagents<br/>ingest · query · lint"]
SKILL["Agent-skill loop · LangGraph<br/>author → eval → gate → human review"]
end
IF --> ING
IF --> REASON
ING --> KG
ING --> VEC
VEC --> RAG
RAG --> EMB["Embeddings · OpenAI"]
RAG <--> MEM
KG <--> AGENT
AGENT <--> MEM
AGENT <--> MCP["External MCP servers<br/>Supabase · GitHub · fetch<br/>reads join agent · writes gated"]
SKILL --> SKL
SKILL --> KG
SKILL --> MCP
REASON --> LLM["LLM providers<br/>Claude · OpenAI · Qwen<br/>DeepSeek · Gemini · Mistral"]
REASON -.->|offline metrics| EVAL["Evaluation<br/>hit@k · MRR · recall<br/>RAGAS · 5-family judge panel"]
REASON -.->|live run traces| OBS["Observability<br/>LangSmith · OpenTelemetry"]
```
Everything above the stores is stateless; all state lives in `data/` (JSON graph,
HNSW vectors, SQLite checkpoints), so the app is reproducible and local-first. The JSON
API can also be exposed *as* an MCP server (reads open, writes deny-by-default).
**Evaluation and observability are different planes — don't conflate thExcerpt of 15,018 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:4e1aa520edbee58c, topic:autonomous-agents, topic:agentic-ai
matched fp:4e1aa520edbee58c, topic:rag, desc:knowledge graph
matched fp:4e1aa520edbee58c, topic:mcp