I built lala because I wanted something that sits between a local LLM runtime and my actual working knowledge, not another generic chatbot, and not a coding agent pretending to know my context. I wanted a tool that could reason over my own notes, docs, research, and project material, stay scoped to a specific project, and run entirely on my machine.

What ships in v1

CLI-first workflow
Project-scoped retrieval
File/folder / RSS ingestion
BM25 / PostgreSQL full-text retrieval
Single-model direct + reasoning flow
lala serve to bootstrap the local runtime

Important: what v1 does not pretend to be

I don’t want to do the usual AI launch thing where the page implies magic that isn’t actually there.

So, plainly:

semantic/vector retrieval is not in the v1 answer path yet
structured memory extraction is not fully real yet
this is not a coding agent
this is not “just works in one click” software yet
There could be many silent bugs present. Please lets me know in case you found

A real setup currently needs:

Docker
an ai-config.yml
local GGUF model files on disk

That setup friction is real, and I’d rather be honest about it than hide it behind marketing copy.

I’d love feedback on 3 things specifically

Does the product category make sense?
Is “local reasoning layer over project knowledge” clear, or is there a better framing?
How painful is the setup story?
I already know this is the weakest part right now. I want blunt feedback on what feels unnecessary or badly designed.
What would make Plan: genuinely valuable for your workflow?
I think this is one of the more interesting parts of lala, but I want to know where it actually helps.

If you try it and it breaks, confuses you, or feels overengineered, say it directly. That’s more useful than polite feedback.

CLI-first workflow
Project-scoped retrieval
File/folder / RSS ingestion
BM25 / PostgreSQL full-text retrieval
Single-model direct + reasoning flow
lala serve to bootstrap the local runtime

Important: what v1 does not pretend to be

I don’t want to do the usual AI launch thing where the page implies magic that isn’t actually there.

So, plainly:

semantic/vector retrieval is not in the v1 answer path yet
structured memory extraction is not fully real yet
this is not a coding agent
this is not “just works in one click” software yet
There could be many silent bugs present. Please lets me know in case you found

A real setup currently needs:

Docker
an ai-config.yml
local GGUF model files on disk

That setup friction is real, and I’d rather be honest about it than hide it behind marketing copy.

I’d love feedback on 3 things specifically

Does the product category make sense?
Is “local reasoning layer over project knowledge” clear, or is there a better framing?
How painful is the setup story?
I already know this is the weakest part right now. I want blunt feedback on what feels unnecessary or badly designed.
What would make Plan: genuinely valuable for your workflow?
I think this is one of the more interesting parts of lala, but I want to know where it actually helps.

If you try it and it breaks, confuses you, or feels overengineered, say it directly. That’s more useful than polite feedback.

lala.ai

Intelligent • Local first • Reasoning ⚡Built for local first

Intelligent • Local first • Reasoning ⚡Built for local first

Important: what v1 does not pretend to be

I’d love feedback on 3 things specifically

Important: what v1 does not pretend to be

I’d love feedback on 3 things specifically