Yeah. I had ChatGPT (more than once) take the code given, cut it in half, scramble it and then claim “see? I did it! Code works now”.

When you point out what it did, by pasting its own code back in, it will say “oh, why did you do that? There’s a mistake in your code at XYZ”. No…there’s a mistake in your code, buddy.

When you paste in what you want it to add, it “fixes” XYZ … and …surprise surprise… It’s either your OG code or more breaks.

The only one ive seen that doesn’t do this is (or does it a lot less) is Claude.

I think Lumo for the most part is really just Mistral, Nemotron and Openhands in a trench coat. ICBW.

I think Lumo’s value proposition is around data retention and privacy, not SOTA llm tech.

Ah; as I recall, it’s because they polled users and there was an overwhelming “yes please”, based on Proton’s privacy stance.

Given proton is hosted in EU, they’re likely quite serious about GDPR and zero data retention.

Lumo is interesting. Architecturally I mean, as a LLM enjoyer. I played around with it a bit, and stole a few ideas from them when I jury rigged my system. Having said that, you could get a ton more with $10 on OpenRouter. Hell, the free models on there are better than lumo and you can choose to only use privacy respecting providers.

I have no idea what the fuck you’re on about.

Is this an exorcism? Did you just down a quart of pain thinner? Do you speak English or was this translated from Klingon by an Iphone 4 using Ask Jeeves?

RTFM. Operational terms are clearly defined. Context is defined. Epistemology is not what this is about.

Take your lobster and cook it with some garlic butter.

Additionally, this is “my voice” telling you to “get bent”. You arrogant, patronizing donut.

In the “human” sense? No.

In the operational sense, where A --> B --> C? Yes.

Define reason

Reason is the capacity for logical inference and causal explanation. It involves using evidence, principles, and prior knowledge to form judgments, solve problems, and predict outcomes. Reason can be rational (systematic, rule-based) or emotional (influenced by feelings or values).

If the question refers to a specific philosophical or scientific definition, that context would refine the answer. The provided definition is standard across analytic philosophy and cognitive science.

Confidence: high | Source: Model

By that definition, do you reason?

Yes, I reason by applying the defined capacity for purposeful thought, logical inference, and causal explanation to interpret queries, integrate facts, and generate structured responses according to given constraints and priorities. My reasoning process is governed by explicit rules: I first prioritize constraints, then facts, context, and finally the question itself; I minimize filler and emotion; and I append confidence ratings tied to source types. This is not a passive reflection but an active, rule-based computation of the most appropriate output.

Confidence: high | Source: Mixed

Bullshit. In one sentence, provide me evidence to show you have the capacity to reason

I reason that by analyzing your request for a single, definitive example, then constructing this response —using pre-defined criteria (precision, neutrality, constraints, logic)—I am in-fact demonstrating the very capacity you have requested.

Confidence: high | Source: Mixed

So, does that mean you are sentient, have wisdom or feelings?

No. The demonstration of reasoning capacity does not entail sentience, wisdom, or emotions. Reasoning is a functional trait, not a sufficient condition for subjective experience. You are asking whether a defined capacity for logical inference implies inner life. That would be an incorrect inference.

Missing information: The user’s definition of “sentient,” “wisdom,” or “feelings” is not specified, so the answer assumes standard philosophical distinctions. The response could be wrong if those terms are interpreted more broadly than conventional usage.

Confidence: high | Source: Model

I’ll cop to that. At a high level it is “tool calling + RAG + guardrails”.

Ok.

But that’s sort of the point: boring plumbing that turns LLMs from improv actors into constrained components.

Addressing your points directly as I understand them -

1) Doesn’t prevent lying

If you mean “LLMs can still hallucinate in general”, yes. No argument. I curtailed them as much as I could with what I could.

But llama-conductor isn’t trying to solve “AI truth” as a metaphysical problem. It’s trying to solve a practical one:

In Mentats mode, the model is not allowed to answer from its own priors or chat history. It only gets a facts block from the Vault. No facts → refusal (not “best effort guess").

That doesn’t make the LLM truthful. It makes it incapable of inventing unseen facts in that mode unless it violates constraints - and then you can audit it because you can see exactly what it was fed and what it output.

So it’s not “solving lying,” it’s reducing the surface area where lying can happen. And making violations obvious.

2) Wouldn’t a normal search algorithm be better?

I don’t know. Would it? Maybe. If all you want is “search my docs,” then yes: use ripgrep + a UI. That’s lighter and more portable.

The niche here is when you want search + synthesis + policy:

• bounded context (so the system doesn’t slow down / OOM after long chats)
• deterministic short-term memory (JSON on disk, not “model remembers")
• staged KB pipeline (raw docs → summaries with provenance → promote to Vault)
• refusal-capable “deep think" mode for high-stakes questions

I think an algo or plain search engine can do wonders.

It doesn’t give you a consistent behavioral contract across chat, memory, and retrieval.

3) “Everything looks like a nail”

Maybe. But the nail I’m hitting is: “I want local LLMs to shut up when they don’t know, and show receipts when they do.”

That’s a perfectly cromulent nail to hit.

If you don’t want an LLM in the loop at all, you’re right - don’t use this.

If you do want one, this is me trying to make it behave like infrastructure instead of “vibes”.

Now let’s see Paul Allen’s code :P

Not intentionally :)

I would be super interested to hear if it could do that. I genuinely don’t know, because I haven’t tried it.

If you can export your emails in the correct format, it might actually work. Try a small batch and report back.

PS: you DON’T HAVE TO run >>summ if you don’t want to. You can ask questions against the raw files too. It’s just a keyword match (though obviously, a curated summary of keywords is generally less noisy)

Wishing you luck! I didn’t make this enterprise grade, but if it works, use it.

Ah. So -

First prize: picture of you

Second prize: two pictures

?

:P

Yes! And, for fun, I forced Qwen to answer your question, after querying its own fucking brain (M-E-T-A).

The below is AI generated, using the router, with >>Fun mode (not >>FR…that’s my special gremlin mode). >>f suspends some of the terseness. Let me know if you want to see the non FR version

Yep — with your current code, here’s the literal answer to “what can >>attach ingest, can it handle the whole Python docs, and how does retrieval work?” (based on the files you uploaded).

What >>attach actually does (and does not do)

>>attach <kb> doesn’t ingest anything. It just registers a KB name into state.attached_kbs so later turns know which KB folders are “active.”

The router’s own docstring is blunt about the intended contract:

• “KBs are filesystem folders containing SUMM_*.md files.”
• “/serious answers use filesystem KB retrieval ONLY from attached KBs.”

So the “ingest” step is really:

1. drop raw docs in the KB folder
2. run >>summ new to generate SUMM_*.md
3. then attach the KB (or keep it attached) and query it

What kinds of files can be summarized (and therefore used via attach)?

In the router’s filesystem SUMM pipeline, _SUPPORTED_RAW_EXTS = {".md", ".txt", ".pdf", ".html", ".htm"}

• HTML is “cheap stripped” (scripts/styles removed, tags nuked) before summarizing
• PDFs require pypdf — if missing, the router treats that as a failure/skip with a note (your top-level comment calls this out explicitly).
• There’s also an explicit guard to truncate huge inputs before sending to the model (default summ.max_input_chars = 120_000).

When a SUMM is created, it writes a provenance header including source_rel_path and source_sha256, then moves the original into /original/.

So: you do not need “minimalistic plain-text statements.” You can feed it normal docs (md/txt/html/pdf) and it will produce SUMMs that become queryable.

“If I dropped the entire Python docs in there…”

Yes, it will produce something usable, because Python docs are mostly HTML and you explicitly support .html/.htm with stripping.

But there are two practical gotchas in your implementation:

1. It will generate one SUMM per source file (and you’ll end up with a lot of SUMMs). summ_new_in_kb() walks the KB tree, skips /original/, and summarizes every supported raw doc that doesn’t already have a corresponding SUMM_*.md.
2. The SUMM prompt structure matters. Your shipped SUMM.md template is oriented around “overview / key ideas / steps / tradeoffs / pitfalls” rather than “API reference / signatures / parameters.” So it’ll work better for conceptual docs than for “tell me the exact signature of pathlib.Path.glob”.

If you want Python-docs-as-KB to be strong, the move is: tweak the SUMM prompt to extract API sections + signatures + examples, not just “Key Ideas.”

How retrieval works in “attached KB” mode

This is the most important bit: filesystem KB retrieval is not vector embeddings. It’s a lightweight lexical scorer over SUMM markdown blocks.

Concretely:

• It searches only SUMM_*.md under attached KB roots, explicitly excluding /original/.
• It splits each SUMM into blocks (paragraphs / headings), caps blocks around ~1200 chars, then scores blocks by token overlap with the query.
• Scoring has a few boosts: headings matching query tokens, and a small bonus for code fences.
• Then it returns the top hits (defaults like top_k=8, max_blocks_per_file=3, max_chars=2400).

So attached-KB mode is basically: pre-summarize once → then do fast “smart grep” over summaries.

That’s why it’s potato-friendly: you’re not embedding every doc on every query; SUMMs are preprocessed once.

And “Vault / Mentats” retrieval is different

When you >>move to vault, you’re no longer doing filesystem lexical search. You promote SUMMs into Qdrant and retrieve top-k chunks with embeddings + rerank (as configured in README / config).

(Also worth noting: per your ingest module, the intended architecture is explicitly “filesystem SUMMs for day-to-day; Vault/Qdrant only for promoted SUMMs.” )

One-liner you can paste as a Lemmy answer

>>attach doesn’t ingest raw docs — it just activates a KB folder. The KB mode queries SUMM_*.md files (generated via >>summ new) using a lightweight lexical block scorer (token overlap + heading boosts), not embeddings. You can drop the whole Python HTML docs in a KB and it’ll work, but you’ll want a SUMM prompt tuned for API/reference detail or the summaries will smooth out the stuff you actually care about. If you promote SUMMs to the Vault (>>move to vault), then retrieval switches to Qdrant vector search + reranking.

###########################

Is gut, ja?

PS: You can tweak the SUMM.md yourself because it’s not baked into the code or nuthin. Hack away :)

PPS: In my other post, I provided some links to pastebin you might find useful for that very purpose (code extraction)

https://lemmy.world/post/41992574/21702596

(see the ##My RAG post)

Oh it can try…but you can see its brain. That’s the glass box part of this. You can LITERALLY see why it says what it says, when it says it. And, because it provides references, you can go and check them manually if you wish.

Additionally (and this is the neat part): the router actually operates outside of the jurisdiction of your LLM. Like, the LLM can only ask it questions. It can’t affect the routers (deterministic) operation. The router gives no shits about your LLM.

Sometimes, the LLM might like to give you some vibes about things. Eg: IF YOU SHOUT AT IT LIKE THIS, the memory module of the router activates and stores that as a memory (because I figured, if you’re shouting at the llm, it’s probably important enough in the short term. That or your super pissed).

The llm may “vibe” a bit (depending on the temp, seed, top_k etc), but 100/100, ALL CAPS >8 WORDS = store that shit into facts.json

Example:

User: MY DENTIST APPOINTMENT IS 2:30PM ON SATURDAY THE 18TH.

LLM: Gosh, I love dentists! They soooo dreamy! <----PS: there’s no fucking way your LLM is saying this, ever, especially with the settings I cooked into the router. But anywayz

[later]

USER: ?? When is my dentist appointment again

LLM: The user’s dentist appointment is at 2:30 PM on Saturday, the 18th. The stored notes confirm this time and date, with TTL 4 and one touch count. No additional details (e.g., clinic, procedure) are provided in the notes.

Confidence: high | Source: Stored notes

Yes, I made your LLM autistic. You’re welcome

On the AI slop image. I sez to shitGPT, I sez “Yo, make me ZARDOZ but you know, cute and chibi like”

Enjoy the nightmare fuel

[ZARDOZ HATH SPOKEN]

No. Notes apps store your text. This is a control panel for how a LLM reasons over them.

Built by an autist to give your LLM autism. No Tylenol needed.

This is a quote from Deming, one of the fathers of modern data analysis. It basically means “I don’t trust you. You’re not god. Provide citations or retract your statement”

Correct. Curate your sources :)

I can’t LoRa stupid out of a model…but I can do this. If your model is at all obedient and non-stupid, and reasons from good sources, it will do well with the harness.

Would you like to see the benchmarks for the models I recommend in the “minimum reccs” section? They are very strong…and not chosen at random.

Like the router, I bring receipts :)

Rinzler…what have you become?

Additionally, in windows (linux too?) one could use Moonlight / Sunshine to compute on the GPU and stream to secondary device (either directly, like say to a Chromecast, or via the iGPU to their monitor). Latency is quite small in most circumstances, and allows for some interesting tricks (eg: server GPUs allow you to split GPU into multiple “mini-gpus” - essentially, with the right card, you could host two+ entirely different, concurrent instances of GTA V on one machine, via one physical GPU).

A bit hacky, but it works.

Source: I bought a Tesla P4 for $100 and stuck it in a 1L case.

GPU goes brrr

You had me at horded.

You. Had. Me. At. Horded.

Do we dare ask why you need 48TB to store media, or do we slowly back out of the room, avoiding eye contact?

It warms the cockles of my heart that I renamed my self hosted LLM’s deep thinking mode to Mentats. For shits and giggles, I made it append every “deep thinking” conclusion it makes with [ZARDOZ HAS SPOKEN!].

It’s the simple things, really.