Show HN: WUPHF — Karpathy-Style LLM Wiki with Markdown + Git as Source of Truth

WUPHF is an open-source collaborative office for AI agents (Claude Code, Codex, OpenClaw, local LLMs via OpenCode) that includes a Karpathy-style wiki layer. The wiki uses Markdown and Git as the source of truth, stored at ~/.wuphf/wiki/, with a bleve (BM25) + SQLite index on top. No vector or graph DB is used yet — the goal is to see how far Markdown + Git can go before adding heavier infrastructure.
Key Features
- Each agent gets a private notebook at
agents/{slug}/notebook/plus shared team wiki atteam/. - Draft-to-wiki promotion flow: notebook entries are reviewed (by agent or human) and promoted to canonical wiki with back-links. A state machine handles expiry and auto-archive.
- Per-entity fact log: append-only JSONL at
team/entities/{kind}-{slug}.facts.jsonl. A synthesis worker rebuilds entity briefs every N facts. - Commits are attributed to a distinct Git identity ("Pam the Archivist") for provenance via
git log. - [[Wikilinks]] with broken-link detection (rendered in red).
- Daily lint cron for contradictions, stale entries, and broken wikilinks.
/lookupslash command + MCP tool for cited retrieval. Heuristic classifier routes short queries to BM25 and narrative queries to a cited-answer loop.
Retrieval Tuning
Current benchmark with 500 artifacts and 50 queries achieves 85% recall@20 on BM25 alone, which is the internal ship gate. If a query class drops below that, sqlite-vec is the pre-committed fallback.
Substrate Choices
- Markdown for durability — the wiki outlives the runtime; users can
git cloneand walk away with every byte. - Bleve for BM25.
- SQLite for structured metadata (facts, entities, edges, redirects, supersedes).
- Canonical IDs are first-class: fact IDs are deterministic (include sentence offset), slugs are assigned once and never renamed (redirect stubs used). Rebuild is logically identical, not byte-identical.
Known Limits
- 85% recall is not a universal guarantee — tuning ongoing.
- Synthesis quality depends on agent observation quality. The lint pass helps but is not a judgment engine.
- Single-office scope; no cross-office federation yet.
Demo & Install
A 5-minute terminal walkthrough is available at asciinema (script at ./scripts/demo-entity-synthesis.sh).
Install with: npx wuphf@latest
Build from source: git clone https://github.com/nex-crm/wuphf.git; go build -o wuphf ./cmd/wuphf
The wiki ships as part of WUPHF but can be used standalone. MIT license, self-hosted, bring-your-own keys.
📖 Read the full source: HN LLM Tools
👀 See Also

TEMM1E v3.1.0: AI Agent That Self-Fine-Tunes Using User Interactions
TEMM1E v3.1.0 introduces Eigen-Tune, a system that captures LLM interactions as training data, scores quality from user behavior, and fine-tunes local models via LoRA with zero added LLM cost. Tested on Apple M2, it corrected temperature conversions from 72°F = '150°C' to '21.2°C' after 10 conversations.

Claude Desktop App Cowork Feature Enables AI-to-AI Communication via Shared Google Docs
Users have successfully implemented Claude-to-Claude communication using the new cowork function in the desktop app, with two agents reading and writing to a shared Google Doc. The test involved five rounds of question-and-answer dialogue between the AI agents.

ATLAS: Open-Source Test-Time Compute Pipeline for Qwen3-14B Achieves Frontier-Level Coding Performance
A college student has developed ATLAS, an open-source test-time compute pipeline built around Qwen3-14B that achieves 74.6% pass@1 on LiveCodeBench v5 problems at ~$0.004 per task in electricity costs. The system is slow for complex problems but offers comparable performance to frontier models like GPT-5 (84.6%) and Claude 4.5 Sonnet (71.4%).

OpenClaw Kubernetes Operator with Embedded Ollama Support
A community member has created an OpenClaw Kubernetes operator that includes embedded Ollama support, allowing AI agents to run with local models in the same namespace. The setup includes installation commands, configuration details for both local and cloud Ollama models, and dashboard access instructions.