Лендинг Trip2G под LLM Wiki

1. Мультиагентность ломает простую схему

AEVYRA описывает failure modes:

• разные CLAUDE.md / CODEX.md / GEMINI.md расходятся;
• агенты создают duplicate concept pages;
• canonical становится самоатрибуцией агента;
• open questions умирают в chat history;
• нет места для honest in-session observations.

Вывод для Trip2G:

Trip2G может быть не просто wiki hosting, а shared coordination layer для многих агентов.

2. Нужны provenance, hashes, audit, deterministic rails

WikiLoom и критические комментарии подчёркивают:

• LLM может тихо испортить wiki;
• summary is lossy compression;
• нужны source hashes;
• нужны span/chunk citations;
• нужны audit logs;
• нужны validators/lint;
• human edits надо защищать;
• нужны rollback/version history.

Вывод для Trip2G:

Не обещать “магическая wiki сама себя пишет”. Обещать “agent-maintained wiki with instructions, logs, provenance and review”.

3. Рынок уже строит похожие продукты

В комментариях упоминаются:

• SwarmVault — compiler, graph, context packs, many agent integrations;
• WikiLoom — deterministic linking/provenance;
• MOM — decision memory and lifecycle;
• Keppi — graph traversal/context packs;
• Kompl — async batch compile;
• llmwiki-cli / Link / ΩmegaWiki / Aura / Lore / Remindb и др.

Вывод:

Это не одинокая идея; это формирующаяся категория. Trip2G должен занять свой угол: MCP-native hosted/networked markdown wiki, где инструкции и pages доступны агентам как tools.

4. Критика: это не замена RAG, а другой слой

Один комментарий справедливо критикует hype:

• нет benchmarks;
• нет scale curve;
• wiki может терять caveats/minority views;
• всё равно нужна retrieval/search на большом масштабе;
• production issues: permissions, audit logs, rollback, sensitive data, concurrency, compliance.

Вывод для лендинга:

Не говорить “RAG is dead”. Говорить “RAG retrieves; LLM Wiki compounds. Use both when needed.”

One-liner

Trip2G turns Karpathy-style LLM Wikis into MCP-accessible knowledge hubs for your agents.

Русская версия:

Trip2G превращает LLM Wiki в MCP-доступную базу знаний, которую агенты могут читать, обновлять, цитировать и связывать с другими базами.

Главный claim

Compile knowledge once. Let every agent reuse it.

Antagonist

Не “RAG плохой”, а:

Your agents keep rediscovering the same context from scratch.

Category

Варианты:

• LLM Wiki hosting for agents;
• MCP-native living knowledge base;
• compounding markdown memory for AI agents;
• hosted LLM Wiki + MCP + automation;
• agent-readable knowledge wiki.

URL

/llm-wiki-for-agents

Внутреннее имя заметки/лендинга:

LLM Wiki for Agents

Hero

Build an LLM Wiki your agents can actually use.

Subtitle

Trip2G turns raw sources, notes and docs into a persistent markdown wiki exposed through MCP — so Claude, Cursor, Codex and other agents stop rediscovering the same context every session.

CTA

Primary:

Create my first LLM Wiki

Secondary:

Connect an existing Obsidian vault

Technical secondary:

Read the agent instructions example

Variant A — developer sharp

Stop feeding agents the same context twice.

Compile your sources into a living markdown wiki. Expose it through MCP. Let every agent search, cite and update the same knowledge base.

Variant B — Karpathy resonance

The wiki is the artifact. Not the chat.

Trip2G gives your LLM Wiki a hosted home, MCP access, sync, permissions and automation.

Variant C — Obsidian/PKM

Obsidian is your IDE. Agents are the maintainers. Trip2G is the MCP hub.

Build a compounding knowledge base from your notes, sources and questions.

Variant D — team/company

Turn scattered team knowledge into a maintained wiki for AI agents.

Meeting notes, docs, Slack exports and research become a structured markdown wiki your agents can query and keep current.

1. Problem

## Your agents keep starting from zero

You upload docs. The model retrieves chunks. It answers. Then the context disappears.

Next session, another agent has to rediscover the same knowledge again.

That works for one-off Q&A. It fails when knowledge should compound.

2. Insight

## An LLM Wiki compiles knowledge once

Instead of querying raw documents every time, an agent builds a persistent markdown wiki:

- source summaries;
- entity pages;
- concept pages;
- comparisons;
- contradictions;
- cross-links;
- index and log.

The wiki gets richer with every source and every good question.

3. Trip2G layer

## Trip2G makes the wiki usable by agents

Trip2G adds the missing operational layer:

- sync markdown from Obsidian/Git/CLI;
- expose the wiki through MCP;
- let agents call `search`, `note_html`, `similar`;
- publish selected pages publicly;
- keep private pages private;
- run change/cron webhooks;
- federate multiple knowledge bases.

4. How it works

## How it works

```text
Raw sources
  ↓ ingest by agent
Structured markdown wiki
  ↓ Trip2G sync
MCP knowledge hub
  ↓ Claude / Cursor / Codex / other agents
Answers with citations + maintained pages

### 5. Query modes

```markdown
## Use RAG, or just follow the index

Trip2G does not force one retrieval strategy.

Small/medium wiki:

1. Read `index.md`.
2. Choose relevant pages.
3. Follow wikilinks.
4. Open only needed sections.

Large wiki:

1. Use Trip2G search / vector / hybrid retrieval.
2. Open source pages.
3. Traverse links and related notes.
4. File valuable answers back into the wiki.

The retrieval behavior is controlled by an instruction note stored next to the knowledge base.

6. Example use cases

## Use it for

- research wikis;
- competitive intelligence;
- codebase/project memory;
- founder/customer discovery notes;
- academic literature review;
- team decision memory;
- book/course companion wikis;
- due diligence;
- personal second brain.

7. Differentiation

## Why not just RAG?

RAG retrieves context.

An LLM Wiki maintains knowledge:

- contradictions are already flagged;
- summaries are already updated;
- links are already created;
- useful answers can become pages;
- agents share the same durable artifact.

Use RAG when you need search. Use the wiki when you need knowledge to compound.

8. Safety/provenance

## Keep it auditable

A good LLM Wiki needs more than summaries.

Trip2G encourages:

- raw sources as source of truth;
- `index.md` and `log.md`;
- source links and hashes;
- protected human edits;
- private/public controls;
- lint passes for broken links, stale claims and contradictions.

9. Demo

## Demo: ask once, keep forever

Ask:

> Compare our positioning against three competitors.

The agent:

1. reads the Trip2G wiki index;
2. opens competitor pages;
3. writes the comparison;
4. saves it as `comparisons/positioning.md`;
5. updates `index.md` and `log.md`.

Next session, any agent can reuse that comparison instead of recomputing it.

10. CTA

## Start with one source

Bring one article, one folder, one Obsidian vault, or one research question.

Trip2G will help turn it into an agent-readable wiki.

CTA:

Create my first LLM Wiki

Free

First wiki / first source free.

Includes:

• one hosted Trip2G base;
• markdown pages;
• basic MCP;
• public/private pages;
• example instructions.

Paid expansion

• more sources;
• more storage;
• scheduled ingest;
• webhooks/cron;
• private MCP endpoint;
• team permissions;
• federation between wikis;
• custom templates;
• provenance/audit tooling;
• managed import from Drive/Notion/Slack/GitHub.

“Isn’t this just RAG?”

No. RAG finds chunks at question time. An LLM Wiki keeps a structured synthesis current over time. Trip2G can use retrieval when useful, but the main artifact is the maintained markdown wiki.

“Won’t the LLM hallucinate into the wiki?”

That is why the wiki needs raw sources, source links, hashes, logs, review, lint and human approval. Trip2G should not promise blind automation; it should support agent-maintained, auditable knowledge.

“What happens when the wiki gets large?”

At small scale, `index.md` and wikilinks are enough. At larger scale, use Trip2G search/RAG plus graph-like traversal. The instruction note tells agents which mode to use.

“Why Trip2G instead of local files only?”

Local files are the foundation. Trip2G adds MCP access, publishing, permissions, webhooks, hosted sync and federation — so multiple agents and people can use the wiki safely.

Hero image prompt

A high-end developer tool landing hero showing a living markdown wiki as a glowing knowledge graph. Raw documents flow in from the left as papers, PDFs, code files and notes. In the center, they compile into structured markdown pages with wikilinks. On the right, multiple AI agents connect through MCP and read only the relevant pages. Calm dark interface, subtle cyan/violet graph lines, local-first and technical aesthetic, no text in image.

Diagram prompt

Create a clean architecture diagram: Raw Sources → Agent Ingest → Structured Markdown Wiki → Trip2G Sync/MCP Hub → Claude/Cursor/Codex. Include optional branches: index-first traversal, RAG search, webhooks/cron, federation. Minimal dark SaaS style.