agentsclimarketplace
Sign in

Llm wiki local first stack

Skill po4yka/llm-wiki-skills/skills/llm-wiki-local-first-stack

Design a local-first LLM-Wiki stack. Use when the user wants Markdown, git, Obsidian, ripgrep, local embeddings, qmd-style retrieval, SQLite/FTS/vector indexes, local LLMs, offline operation, or safe sync without cloud lock-in; route retrieval/index layer design to llm-wiki-retrieval-architect.From its SKILL.md

Install
npx -y skills add po4yka/llm-wiki-skills --skill llm-wiki-local-first-stack

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.2 KB, 580 tokens by cl100k_base, as published. Nobody here has run it

LLM-Wiki Local-First Stack

Goal

Help the user choose and configure a durable local-first LLM-Wiki stack with the least necessary infrastructure.

When to use

  • The user wants their LLM-Wiki to run offline or with minimal cloud dependency.
  • The user asks whether they need a vector database, graph store, or SQLite/FTS index yet.
  • The user is deciding between Markdown+git+rg and a heavier retrieval or product storage layer.
  • The user wants a sync strategy for a vault shared across multiple devices without breaking index files.
  • The user needs a model policy split between local-only and cloud-assisted tasks for privacy reasons.

Inputs

  • Corpus size and growth rate.
  • Privacy/offline requirements.
  • Existing tools: Obsidian, git, qmd, SQLite, vector DB, local LLMs.
  • Target machine constraints.
  • Desired agent integration.

Procedure

1. Start with the minimum viable stack

Default:

Markdown + git + index.md + log.md + rg + Agent Skills

Do not add vector databases or graph stores unless symptoms justify them.

2. Choose retrieval tier

TierUse when
index.md + rgEarly and medium vaults with good titles and wikilinks.
hybrid local searchExact search misses conceptual matches.
graph-aware retrievalRelationship and multi-hop questions dominate.
product storageConcurrency, permissions or scale require it.

3. Decide storage policy

Prefer:

  • Markdown as source of truth;
  • SQLite/FTS for local indexes;
  • reconstructable vector indexes;
  • content hashes for incremental rebuild;
  • git for text and manifest files;
  • per-device generated indexes.

Avoid syncing mutable DB/index files unless the user has a tested sync strategy.

4. Decide model policy

Classify tasks:

  • local-only ingest for sensitive sources;
  • cloud-assisted query for public material;
  • cheap model for triage;
  • stronger model for synthesis;
  • local embeddings/reranking when privacy matters.

Hand off to llm-wiki-model-policy for detailed policy.

5. Produce setup plan

Include:

  • folder structure;
  • agent instructions;
  • retrieval tier;
  • index/cache policy;
  • backup/sync plan;
  • upgrade triggers.

Output

## Local-first recommendation

## Minimal stack

## Retrieval tier

## Storage and sync policy

## Model policy

## Setup steps

## Upgrade triggers

Safety gates

  • Do not recommend syncing non-mergeable index files without warning.
  • Do not make cloud services mandatory for local-first users.
  • Browse before giving current install commands.
  • Keep raw sources portable and human-readable where possible.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,144. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.