Cairni
Overview
Cairni is a managed LLM Wiki service that takes Andrej Karpathy's LLM Wiki pattern and runs it as a hosted product. Where the self-hosted pattern requires a user to assemble their own agent, schema file, search tooling, and backup infrastructure, Cairni handles all of that automatically — described by its own positioning as *"zero setup."* llm-wiki.en.md
Core Idea
Cairni's design premise directly mirrors Karpathy's mental model: your knowledge is the codebase, and the AI is the programmer. Cairni serves as that programmer in hosted form, so the wiki stays maintained because the cost of maintenance is near zero. llm-wiki.en.md
Users drop in sources — files, URLs, or even meeting recordings (transcribed with speaker separation) — and the AI compiles them into a living, interlinked wiki, with every claim traceable back to its source. llm-wiki.en.md
Features
- Zero setup — no agent configuration, schema files, or tooling to assemble manually
- Source traceability — every claim links back to the raw immutable source it came from
- Scale beyond ~1,000 files — addresses the ceiling of the self-hosted LLM Wiki index-first approach
- Team workspaces — per-notebook access control and approval queues for multi-user collaboration
- Version history — tracks how the wiki evolves over time
- Contradiction surfacing — flags conflicting claims across pages
- Public publishing — wikis can be published publicly llm-wiki.en.md
How It Closes the Self-Hosted Gap
The self-hosted LLM Wiki pattern has several well-known limitations that Cairni is positioned to address:
| Limitation | Self-Hosted | Cairni |
|---|---|---|
| Scale | ~1,000 files comfortably | Beyond the ~1,000-file wall |
| Multi-user | Single-player by default | Team workspaces with access control |
| Setup | Assemble agent, schema, search, backups yourself | Zero setup |
| Governance | No built-in confidence scoring or review | Approval queues |
llm-wiki.en.md
Relationship to the LLM Wiki Pattern
Cairni is described as an implementation of the open LLM Wiki pattern, not a departure from it. The personal and team versions share the same underlying model: *"personal = a team of one."* The Ingest / Query / Lint Workflow, the compiled-over-retrieved philosophy, and the distinction from RAG (Retrieval-Augmented Generation) all carry over directly from the open pattern. llm-wiki.en.md
Caveats
The source describing Cairni is a draft written for Cairni's own publishing platform (Cairni Explore), which means it should be treated as self-promotional material. Claims about specific capabilities (e.g., scale ceiling removal, speaker-separated transcription) are taken from that source without independent verification. llm-wiki.en.md