Show HN: 10x better performance from the Coding Harnesses with LLM-wiki (opens in new tab)

(llm-wiki.net)

16 pointsnvk9d ago14 comments

14 comments

14 comments · 7 top-level

esafak9d ago· 3 in thread

How does this differ from https://context7.com/ ?

LLM Wiki is client-side and local-first (plain Markdown, Obsidian-friendly) designed for deep multi-agent topic research (e.g. automated thesis/counter-thesis runs, local session memory redaction).

Context7 is a hosted SaaS/on-premise MCP server indexing API/library docs (GitHub, Confluence, OpenAPI) to provide coding assistants with fresh, version-specific developer context.

Essentially: LLM Wiki compiles topic research vaults on your local disk, whereas Context7 acts as a semantic doc/API search gateway for programming.

esafak9d ago

So the benefit is in caching the resources to avoid web queries, and massaging them to make them amenable to analysis? For intellectual work I imagine it would be useful if it could access gated content, like commercial reports?

kordlessagain9d ago

Right, it's the whole search pipeline problem defined. Check this out: https://github.com/deepbluedynamics/lume

jarbus9d ago· 2 in thread

Where is this 10x number coming from?

dominotw9d ago

from llm wiki

esafak9d ago

The 10x coder who wrote it: https://www.youtube.com/watch?v=d2BuP7-m5Ww

xnx9d ago· 1 in thread

Some say text files work even better: https://cloud.google.com/blog/products/data-analytics/how-th...

eikenberry9d ago

This makes total sense given that a wiki is a collection of plain text files with a few simple conventions (not markup) around formatting and linking.

adithyassekhar9d ago· 1 in thread

Edit: On closer look this looks useful without the coding harness pitch.

nvkOP9d ago

This is not a product, its a foss lib

superfrank9d ago

I've got a couple of LLM wikis running for different purposes. I just pointed Claude at Karpathy's Github Gist and said "do this" and it set up and has maintained them ever since. So far no issue with that.

Can you explain why the version linked is better?

Gist link: https://gist.github.com/karpathy/442a6bf555914893e9891c11519...

dolebirchwood9d ago

What if I only want 6x better performance? Is there a knob or slider to dial it down?

sppfly9d ago

This kind of llm bragging title and AI generated webpage makes me gross.

j / k navigate · click thread line to collapse

14 comments

14 comments · 7 top-level

esafak9d ago· 3 in thread

How does this differ from https://context7.com/ ?

kordlessagain9d ago

LLM Wiki is client-side and local-first (plain Markdown, Obsidian-friendly) designed for deep multi-agent topic research (e.g. automated thesis/counter-thesis runs, local session memory redaction).

Context7 is a hosted SaaS/on-premise MCP server indexing API/library docs (GitHub, Confluence, OpenAPI) to provide coding assistants with fresh, version-specific developer context.

Essentially: LLM Wiki compiles topic research vaults on your local disk, whereas Context7 acts as a semantic doc/API search gateway for programming.

esafak9d ago

kordlessagain9d ago

Right, it's the whole search pipeline problem defined. Check this out: https://github.com/deepbluedynamics/lume

jarbus9d ago· 2 in thread

Where is this 10x number coming from?

dominotw9d ago

from llm wiki

esafak9d ago

The 10x coder who wrote it: https://www.youtube.com/watch?v=d2BuP7-m5Ww

xnx9d ago· 1 in thread

Some say text files work even better: https://cloud.google.com/blog/products/data-analytics/how-th...

eikenberry9d ago

This makes total sense given that a wiki is a collection of plain text files with a few simple conventions (not markup) around formatting and linking.

adithyassekhar9d ago· 1 in thread

Edit: On closer look this looks useful without the coding harness pitch.

nvkOP9d ago

This is not a product, its a foss lib

superfrank9d ago

Can you explain why the version linked is better?

Gist link: https://gist.github.com/karpathy/442a6bf555914893e9891c11519...

dolebirchwood9d ago

What if I only want 6x better performance? Is there a knob or slider to dial it down?

sppfly9d ago

This kind of llm bragging title and AI generated webpage makes me gross.

j / k navigate · click thread line to collapse