An open-source AI skill that turns every conversation into a permanent, compounding investment-research wiki.
┌─────────────┐
│ YOUR EDGE │
│ COMPOUNDS │
└──────▲──────┘
│
───► earnings ──┐ │
───► news ─┤ │
───► filings ─┼──► AI agent ──────┤
───► tweets ─┤ ▲ │
───► research ─┘ │ │
│ │
OpenOrder skill │
│ │
┌──────▼─────┐ │
│ WIKI │──────┘
│ ~/openorder│
│ (markdown) │
└─────────────┘
Every investor lives the same loop:
Read a Goldman note. Skim 5 tweets. Listen to a podcast. Watch an earnings call. Forget 90% of it by next week. Repeat.
Most "AI for research" tools push you deeper into that loop — you upload PDFs to NotebookLM, you ask ChatGPT, the answer is great, then it evaporates into chat history. Next week you're re-deriving the same thesis from the same documents.
That's not research. That's expensive forgetting.
OpenOrder makes your AI agent maintain a persistent wiki about your coverage universe — companies, industries, frameworks, earnings, theses — and update it on every conversation.
You: "What did MSFT just say about capex?"
Agent: [Reads INDEX.md → companies/MSFT.md → answers]
[Detects new earnings → updates MSFT profile]
[Appends to log.md: "2026-05-08 ingest | MSFT FY26 Q3 capex guide"]
You: "And how does that affect NVDA?"
Agent: [Already knows from your wiki that NVDA is 25% of MSFT's
AI capex → cross-references → updates both files]
Your knowledge compounds. The wiki gets richer every conversation. You stop asking the same questions twice.
"The wiki is a persistent, compounding artifact. The cross-references are already there. The contradictions have already been flagged. The synthesis already reflects everything you've read." — Andrej Karpathy, LLM Wiki gist
OpenOrder is a production-ready implementation of that pattern — specialized for investment research and pre-wired to work with every major AI coding agent.
git clone https://github.com/realnaka/OpenOrder.git
cd OpenOrder
./install.shThat's it. The installer:
- Detects which AI agents you have installed (Claude Code, Cursor, Codex, Hermes, OpenCode, OpenClaw)
- Symlinks the skill into each agent's expected location
- Initializes an empty wiki at
~/openorder/withINDEX.md,log.md,raw/, and templates - Sets up local
gitfor free version history
Now open any AI agent and say:
"What's NVDA trading at and what's the latest thesis?"
The skill auto-triggers. The agent reads your wiki, answers, and writes back any new insights.
| Operation | When it runs | What happens |
|---|---|---|
| READ | Every ticker / industry / framework mention | Agent reads INDEX.md then drills into relevant files before answering |
| INGEST | You paste a URL / earnings / tweet | 5-step flow: fetch full text → store in raw/ → extract entities → update wiki → append log.md |
| WRITE | New insight emerges in conversation | Agent updates company profile / framework / portfolio + timestamps everything |
| LOG | Every write | Append-only log.md entry → grep "^## \[" log.md shows your timeline |
| LINT | You say "lint wiki" or monthly | Health-check report: contradictions, stale data, orphan files, missing pages, data gaps |
See SKILL.md for the full specification.
~/openorder/
│
├── raw/ ← Layer 1: immutable sources
│ ├── earnings/ Earnings transcripts, 8-Ks
│ ├── articles/ Tweets, blog posts, research notes
│ ├── filings/ SEC PDFs
│ └── research-notes/ Raw scratch from conversations
│
├── INDEX.md ← Layer 2: your living wiki
├── log.md (LLM-owned, continuously maintained)
├── companies/{TICKER}.md
├── industries/{NAME}/
├── frameworks/{NAME}.md
├── earnings/{TICKER}-{Q}.md
├── portfolios/{NAME}.md
└── templates/
~/.claude/skills/openorder/ ← Layer 3: the schema (this repo)
└── SKILL.md
Why three layers?
- Raw is sacred: you can always re-derive the wiki from raw sources. Raw never changes.
- Wiki is fluid: gets rewritten as understanding evolves.
- Schema is portable: install in 5 seconds across any agent.
OpenOrder is agent-agnostic. The same skill, the same wiki, the same data — across every tool you use.
| Agent | Mechanism | Auto-detected by install.sh |
|---|---|---|
| Claude Code (Anthropic) | SKILL.md (native) |
✅ |
| Cursor (IDE + CLI) | SKILL.md (compatible) |
✅ |
| Codex CLI (OpenAI) | SKILL.md + AGENTS.md |
✅ |
| Hermes Agent (NousResearch) | AGENTS.md + ~/.hermes/skills/ |
✅ |
| OpenCode (sst) | AGENTS.md |
✅ |
| OpenClaw | AGENTS.md |
✅ |
| Aider | CONVENTIONS.md symlink (per-project) |
|
| Cline / Roo Code | .clinerules (per-project) |
|
Anything else supporting AGENTS.md |
AGENTS.md |
✅ via ~/.config/<agent>/ |
Full setup details: docs/compatibility.md.
| OpenOrder | RAG / NotebookLM | |
|---|---|---|
| Knowledge accumulation | ✅ Wiki compounds every session | ❌ Re-derived every query |
| Cross-source synthesis | ✅ Pre-computed, stored in wiki | ❌ Re-discovered each time |
| Contradiction tracking | ✅ Flagged in wiki, logged | ❌ Lost between sessions |
| Evolving thesis | ✅ Bull/bear updated as you learn | ❌ Stuck at upload time |
| Cross-agent portable | ✅ Same wiki across Claude/Cursor/Codex/etc | ❌ Locked to one product |
| Version history | ✅ git out of the box | ❌ Black box |
| Privacy | ✅ 100% local markdown | |
| Cost | ✅ Free (your existing agent's API key) | 💲 Per-product subscription |
- Personal investor — track a 30-stock coverage universe with bull/bear theses that update as earnings drop
- Buy-side analyst — maintain a "structural shorts" wiki with sourced bear cases and conviction scores
- Sell-side researcher — keep a living industry map with peer comparison tables that auto-refresh
- Crypto researcher — track 50 protocols with TVL trends, governance changes, and tokenomics edits
- Family office / investment club — share a wiki across teammates via a private GitHub repo
OpenOrder ships investment-research as the default domain, but the architecture is general. To adapt:
- Edit
SKILL.mdSection 3 to swap the trigger keywords for your field - Edit
examples/company-template.mdto match your entity type (drug pipeline / property / DeFi protocol / academic paper) - Edit
examples/INDEX.example.mdto match your taxonomy
See docs/customize.md for adapted examples (crypto, biotech, real estate, academic literature).
OpenOrder ships only the skill + templates. It does not include:
- Real research data (your wiki is local-only at
~/openorder/, never committed here) - A specific opinion on any company or industry
- Telemetry, analytics, or network calls
- A specific scoring framework — just an abstract scaffold; bring your own methodology
-
examples/for non-investment domains (crypto, biotech) - Optional
git-synchook for multi-machine / team collaboration - Companion VS Code / Obsidian plugin for visual wiki browsing
- Lint command CLI (
openorder lint) outside the agent - CI template for keeping wiki fresh against earnings calendars
PRs welcome. See AGENTS.md for contributor / agent guidance.
OpenOrder's three-layer architecture, log.md timeline, Lint operation, and Ingest workflow are direct implementations of Andrej Karpathy's LLM Wiki pattern (2026).
The investment-research framing, cross-agent install scaffolding, and packaged templates are original to OpenOrder.
The pattern itself dates back to Vannevar Bush's Memex (1945) — a personal, curated knowledge store with associative trails between documents. Bush couldn't solve the maintenance problem. LLMs can.
MIT. See LICENSE.