cpu-llm

A CPU epistemic text engine. It answers from paragraphs it has already stored.

One question in. A sentence rebuilt from a stored proposition and checked against the retrieved paragraphs. No decoder, no API key, no GPU.

Author: Amin Parva, independent engineer · Retrieval plane: VectorPrism

---

What cpu-llm is

cpu-llm is not a quantized chat model. A chat model writes the next token. cpu-llm retrieves a stored paragraph, rebuilds one sentence from a proposition cut out of that paragraph, and says that sentence only when the rebuild matches the stored text and the auditor accepts it.

The encoder is VectorPrism, frozen all-mpnet-base-v2 plus the projection checkpoint, writing one 1024-dimensional tensor per paragraph into PostgreSQL with pgvector. Deliberation is PrismCognition. The turn graph is ChorusGraph. The sentence itself comes from the realizer in this package. If a paragraph does not round-trip, it stays in the index as text and is not given an invented triple.

stored paragraphs → VectorPrism 1024-d index
                 → cpu-llm ask
                 → checked sentence, plus any extra section the question asked for

prompt → GPU chat model → free-form tokens
prompt → “CPU LLM” GGUF decode

The second picture is a different product. There is no Gemini client, no OpenAI client, and no neural decoder in this package.

---

What a question does

prompt
  → PrismGuard
  → VectorPrism dense shortlist
  → PrismResonance on the dense core
  → six channel readings of that shortlist
  → PrismThinker
  → PrismCognition, the realizer, and the entailment auditor
  → PrismShine
  → optional agents, in this order: deep search, creative writing, comparison, brief
  → reply
  → discovery only when ask(..., discover=True)

A blocked prompt stops at the reply. Think goes straight to cognize. Imagine is not inserted between them.

The checked sentence is answer. The reply prints that sentence first. Extra sections are added only when the question asked for them, in this order: summary or explanation or list or sources, then comparison, then an imagined passage, then deep-search possibilities. Preface, follow-ups, and next steps are not spoken.

Question	What runs	What the reply adds
What is the European lobster?	The fixed chain	The checked sentence only
Deep search the European lobster.	Every tagged paragraph of the seeds, plus relation-hand neighbors	A possibility list. The checked sentence stays
Imagine the European lobster.	Stored sentences from the article, a different subset each run	An imagined passage after the checked sentence
Compare the European lobster and Robert Boulter.	Each side's own stored article or tagged paragraphs	A comparison section. No invented "X differs from Y" sentence
Summarize the European lobster.	PrismCognition ranks the stored sentences	Up to three sentences. A contradiction withholds the summary
Why is the European lobster fished with pots?	Only stored sentences that state a cause	An explanation, or "No stored sentence states a cause."
List the European lobster.	The same ranking, kept longer	Up to eight sentences
Cite sources for the European lobster.	The paragraph ids that were used	A source list

Those agents chain. "Deep search and compare the European lobster and Robert Boulter" gathers the possibilities first, then splits them by topic. "Summarize that" with no new topic words reuses the previous turn's paragraph ids. A normal follow-up such as "what about the claws?" does not.

Discovery stays off until discover=True. A why question does not turn it on.

---

The default corpus

Two sources ship inside the package.

WikiText-2 raw, the Salesforce validation and test splits. They are English Wikipedia articles that editors marked Good or Featured. The validation split is 60 articles and 1,841 paragraphs, stored under wikitext2-. The test split is a second set of articles with no shared title, stored under wikitext2test-. A proposition is stored only when the realizer can repeat the paragraph's first sentence. The rest stay text. License of that text: CC BY-SA 3.0 and GFDL, as stated in the provenance file next to the parquet.

An original primer on artificial intelligence, src/cpu_llm/data/artificial-intelligence.md, 70 paragraphs. It is text written for this package. It is not a copy of another textbook. Each paragraph's first sentence is one the realizer can repeat, so a question about search, learning, a transformer, or cpu-llm itself can be answered from those paragraphs. Its ids start with artificial-intelligence-md-, so loading it does not delete WikiText.

cpu-llm-ingest

That command loads the WikiText validation split and then the primer. The same entry point is python -m cpu_llm.ingest.

Command	What it writes
cpu-llm-ingest	WikiText validation, then the primer
cpu-llm-ingest --textbook	The primer only. Use this when WikiText is already in the index
cpu-llm-ingest --test	The WikiText test split only

---

Install

Python 3.11 or newer. The GitHub repository is public. The package name on PyPI is cpu-llm. Install from a checkout:

git clone https://github.com/insightitsGit/cpullm.git
cd cpullm
python -m venv .venv
.venv\Scripts\activate
pip install -U pip
pip install -e .

pip install -e . installs the package, the trained checkpoint, the WikiText parquet files, and the primer. On Windows, import pyarrow before torch or VectorPrism. cpu_llm.retrieval already does that.

The appliance talks to PostgreSQL with the vector extension:

image    pgvector/pgvector:pg16
port     localhost:5433
user     vectorprism
password vectorprism
database vectorprism

postgresql://vectorprism:vectorprism@localhost:5433/vectorprism

VECTORPRISM_PG_DSN overrides that DSN.

Pinned libraries in pyproject.toml:

Package	Role
vectorprism 0.1.3	1024-d encode and dense-slice search
chorusgraph 1.3.0	Turn graph
prismguard 0.1.10	Prompt check
prismcognition 2.1.0	Deliberation and summary ranking
prismrag-patch 0.2.1	Article label from the document id
prismresonance 0.3.0	Dense-core resonance
prismcortex 0.4.1	Graph memory behind the sqlite store
prismmanifest 0.3.4	Numeric span grammar on the CPU
prismshine 0.2.2	Retrieval and sentence gate, judge off
prismthinker 1.2.0	Sufficiency: answer, gather, refuse, escalate
prismagenticpay 1.4.1	Payment check when both a proposal and an authority are passed

A clean install works when those pins can be downloaded from PyPI. PrismThinker 1.2.0 and PrismAgenticPay 1.4.1 are both published. PrismAgenticPay 1.4.1 requires PrismThinker 1.2.0, and both are imported at startup.

---

Runtime

cpu-llm runs on a CPU. It does not use a GPU. Startup clears CUDA_VISIBLE_DEVICES. If PyTorch was already imported and CUDA is available, the process refuses to continue.

Requirement	This release
Python	3.11 or newer
Processor	A 64-bit CPU. No CUDA device
Encoder	sentence-transformers/all-mpnet-base-v2, about 418 MiB, downloaded on first use and run on the CPU
Checkpoint	vectorprism.pt, about 3.6 MB, shipped inside the package
Database	PostgreSQL 16 with the vector extension
Questions	One ask at a time on the calling thread

The five PrismThinker heads are separate processes, started once when the appliance loads. They do not let two questions run together.

Measured latency

These times are from one machine on 13 October 2026: an Intel Core Ultra 9 275HX with 24 cores and 64 GB of RAM, Windows 11, Python 3.12. One appliance, three runs of each prompt, after the model was already loaded and after one warmup question. Load time is not included in the question times.

Starting the appliance took 9.8 seconds on the six-mode run and 10.2 seconds on the deep-search rerun. The warmup question took 233 ms.

Mode	Prompt	Median	Three runs
Normal	What is the European lobster?	66 ms	83, 56, 66
Explain	Why is the European lobster fished with pots?	60 ms	87, 60, 59
Summarize	Summarize the European lobster.	63 ms	72, 60, 63
Compare	Compare the European lobster and Robert Boulter.	61 ms	87, 61, 61
Creative	Imagine the European lobster.	66 ms	67, 66, 66
Deep search	Deep search the European lobster.	593 ms	577, 593, 629

Deep search kept 582 possibilities and wrote a 32,639-character reply. Creative writing returned a different passage on each run. Ingesting a two-sentence note took 4.50 s, then 2.07 s, then 2.40 s, because each paragraph is encoded. A later upload of a 9-paragraph PDF stored every paragraph and answered from that file.

This machine is the one that was measured. It is not a minimum specification.

Next release

The next release is working on serving multiple threads, so one process can answer concurrent questions under heavy load. This release keeps each question on the calling thread.

---

Quickstart

Load the shipped sources, then ask.

cpu-llm-ingest

from cpu_llm.pipeline import Appliance

with Appliance() as appliance:
    result = appliance.ask("What is artificial intelligence?")

print(result["answer"])
print(result["reply"]["text"])

answer is the checked sentence. Shine and the auditor see answer only.

Index your own source

A source is a file, a folder, or paragraphs you already split. ingest stores those paragraphs in the same index, and a later ask reads them with the WikiText rows and the primer. Accepted files are .txt, .md, .html, .pdf, .docx, .csv, .xlsx, .xlsm, and .epub. A legacy .doc or .xls file is refused. Save it as .docx or .xlsx.

from cpu_llm.documents import SourceDocument
from cpu_llm.pipeline import Appliance

guide = SourceDocument(
    "field-guide",
    (
        "The European lobster is Homarus gammarus.",
        "Homarus gammarus is a species of clawed lobster.",
    ),
)

with Appliance() as appliance:
    appliance.ingest(guide)
    appliance.ingest("notes/lobster.pdf")
    appliance.ingest("notes")
    result = appliance.ask("What is the European lobster?")

Reading the same source again replaces that source's paragraphs. A spreadsheet with from and to columns stores each row as a paragraph and stores the pair as a connection. Other sheets become column: value rows and do not invent a connection.

catalog_chunks is the lower-level call when you already have (document_id, paragraph) pairs. Pass propositions= when you already have an EIRProposition whose subject and object are slices of the chunk. ask(..., propositions=...) overrides that turn. The normal path reads the proposition stored with the chunk.

Leave ObjectiveSpec, policy rules, and causal edges empty unless you are supplying them. The appliance does not invent them from prose.

Tests

python -m pytest

---

How the checked sentence is chosen

VectorPrism searches once. The six channel nodes read that shortlist. They do not run six more database queries.

Channel	What the node reports	Used to pick the passage
Dense	Cosine on the trained dense core	Yes
Relational	A stored proposition that shares the prompt's words	No
Disentangled	Words that appear in only one passage	No
Hyperbolic	Poincaré distance, inside the open unit ball	No
Identity	Names and title tokens shared with the prompt	No
Causal	A sentence that already says because, due to, caused, or therefore	No

A quoted cause is not turned into a graph edge. The detail node prints all six. It does not replace the checked sentence.

The current checkpoint trained the dense channel. Relational, disentangled, hyperbolic, and causal heads are still at initialization. cpu-llm does not pretend those slices rank the search.

PrismThinker can return answer, gather, refuse, or escalate. A gather result still realizes the grounded sentence when a proposition exists. The open questions stay on gather_questions.

Creative writing runs when the prompt says imagine, what if, story, poem, fiction, or creative writing, or when ask(..., imagine=True). It realizes stored sentences from the article and draws a different subset each time more than one sentence passed the auditor. One eligible sentence cannot vary. answer stays the checked sentence.

---

Numeric spans

amount_min is set only from a money span.

Written span	Recorded as
4,399 m	Length in metres. Not money
14,440 feet, 14,440-foot	Length in feet
10 million, exceeds 10 million, five million	amount_min = 10000000 or 5000000
$5 m, £ 1.7 million, CZK 8 million, $ 2.5 million in grants	Money, scaled by one million
28 million years, 15 million copies, 20 million men	Not money. The noun is what was counted
a viewership of 8.69 million	Not money. The noun before of is what was counted
4,399 m and $ 10 million in one paragraph	Length and money, both kept
. m	Ignored. One bad paragraph does not drop the rest of the batch

A height in the retrieved text can still select PrismThinker's empirical regime. The questions no numeric threshold on hypothesis or objective and causal_graph is required mean the caller did not supply those inputs. They are not a request for cpu-llm to invent a target.

---

What this package does not do

It does not rewrite a passage in a new style, translate it, write code, or pick a winner the store does not state. It does not expose an HTTP upload API or a per-customer tenant. The tenant on this appliance is cpu-llm, and every source shares one pgvector index. Concurrent questions are the work of the next release.

A paragraph that does not round-trip is stored and can be retrieved. It is not spoken.

---

Layout

src/cpu_llm/pipeline.py                         turn graph and Appliance.ask / ingest
src/cpu_llm/checkpoints/vectorprism.pt          trained projection shipped in the package
src/cpu_llm/ingest.py                           WikiText and the primer loader
src/cpu_llm/data/wikitext-2-raw-v1/             shipped WikiText parquet files
src/cpu_llm/data/artificial-intelligence.md    shipped primer
src/cpu_llm/documents.py                        file, folder, and spreadsheet readers
src/cpu_llm/source.py                           paragraph cuts for the shipped sources
src/cpu_llm/clauses.py                          first-sentence propositions
src/cpu_llm/realizer/                           controlled English from a proposition
src/cpu_llm/audit/                              lexical entailment check
src/cpu_llm/cognition/                          deliberation adapter
src/cpu_llm/agents/                             deep search, creative writing, reply
src/cpu_llm/comparison.py                       two-topic split
src/cpu_llm/classifier.py                       which optional agent the prompt asked for
src/cpu_llm/measurements.py                     lengths, money spans, qualifiers
src/cpu_llm/channels.py                         six readings and the detail template
docs/cpu-llm-v1.6.1-architecture-spec.md

---

Citation

@software{cpullm2026,
  title  = {cpu-llm: A CPU Epistemic Text Engine on VectorPrism},
  author = {Amin Parva, independent engineer},
  year   = {2026},
  url    = {https://github.com/insightitsGit/cpullm}
}

---

Links

• Author: Amin Parva, independent engineer (parvaamin@gmail.com)
• This repository: https://github.com/insightitsGit/cpullm
• Retrieval engine: https://github.com/insightitsGit/VectorPrism