Prokopton42

How does a text become a conversation?

Before a thinker can answer, we need to find their words, establish what they could have known and retrieve passages relevant to your question. Here is how Prokopton42 does that.

The same steps, at your preferred depth.

Prepare the texts and the thinkers

This happens before conversations begin, and again whenever the corpus, our collection of texts, is expanded or corrected. It produces two things: a searchable library and a historical frame for each thinker.

Step 1 of 5
Passage followed in this exampleSeneca · Letters · 13.4Editions retained
Starting with Different documents
  • A Latin edition
  • A French translation
  • An English translation
Preserve provenance
Result Each source stays identifiable
  • Work and edition
  • Translator, language, origin
  • Original copy retained
Illustrative walkthroughThree editions of a letter by Seneca remain three distinct sources.

1. Keep the text and its history together

We collect works, translations and accounts of thinkers’ lives. Each edition retains its source and its translator or editor where known. The document we received is kept intact, so we can always return to what was actually read.

Under the hood

Source files are archived with provenance, an acquisition date and a digital fingerprint. The fingerprint makes a changed file detectable. Processing creates derived data without rewriting the original. An account of a thinker stays attributed to the person who reported it.

Passage followed in this exampleSeneca · Letters · 13.4Versions linked
Starting with Several ways to read it
  • Latin · letter 13, § 4
  • French · letter 13, § 4
  • English · letter 13, § 4
Link the versions
Result One passage, several languages
  • Seneca · Moral Letters
  • Shared reference: 13.4
  • Each translation keeps its attribution
Illustrative walkthroughA translator’s note must never turn into a sentence attributed to Seneca.

2. Give every passage a stable address

The author’s text is separated from notes and commentary. Passages receive a stable reference, shared across languages where their divisions match. Page numbers change between editions; “letter 13, paragraph 4” still points to the same passage.

Under the hood

This is canonical identity: thinker, work and reference identify a passage independently of an edition. Each translation is attached as a distinct version. Alignment checks these correspondences. Unrecognised structure or incompatible numbering blocks admission to retrieval instead of creating a false match.

Passage followed in this exampleSeneca · Letters · 13.4Historically admitted source
Starting with The project can study…
  • Seneca’s own writings
  • A present-day commentary
  • Marcus Aurelius’ Meditations
Separate the periods
Result What Seneca may claim to know
  • His admitted historical sources
  • Chronological cutoff: 65 CE
  • No Marcus Aurelius or modern knowledge
Illustrative walkthroughA book that helps us study Seneca is not necessarily a book Seneca knew.

3. Establish what this thinker could know

Owning a text does not mean we can put it in someone’s mouth. We distinguish their writings, what they may have read or heard, accounts of their life and later commentary. Their period, ideas and documented relationships set the boundaries of their responses.

Under the hood

A knowledge envelope specifies admitted sources, a chronological cutoff and exclusions for each person. Voice rules are kept separate from biographical facts. Concepts and relationships form a derived, versioned graph: each connection retains its type, direction, evidence and confidence. A mention proves neither reading nor a personal encounter.

Passage followed in this exampleSeneca · Letters · 13.4Retrievable passage
Starting with An admitted passage
  • Exact text
  • Reference 13.4
  • Language and provenance
Prepare semantic search
Result A retrieval index
  • Segments linked to the passage
  • Numerical representations
  • A route back to the exact text
Illustrative walkthroughSimilarity helps find a candidate. It does not yet prove the passage answers the question.

4. Make passages searchable by meaning

Once text, references and languages have been checked, admitted passages are prepared for retrieval. A question about fearing tomorrow can then find a passage about anxiety, even when it uses different words.

Under the hood

Passages are divided into retrieval segments, or chunks, while keeping their references. A local model calculates an embedding for each segment: a list of numbers representing features of its meaning. Qdrant, the vector search engine, compares these representations. The index can be rebuilt; citations always resolve back to the canonical text.

Passage followed in this exampleSeneca · Letters · 13.4Readable passage
Starting with The prepared resources
  • A library of passages
  • Each thinker’s historical frame
  • Sourced concepts and relationships
Enable admitted uses
Result Separate ways to explore
  • Read in the Library
  • Explore connections in the graph
  • Converse when the voice is ready
Illustrative walkthroughAdding a work restarts its preparation. It does not automatically change every thinker’s knowledge.

5. Separate reading, searching and dialogue

A text may be readable in the Library without being usable in an answer. Likewise, a thinker may have a profile without being available in chat. Reading, retrieval and dialogue each have their own opening conditions.

Under the hood

Availability is managed by capability. Corpus and alignment checks come before indexing; dialogue also requires the person’s configuration and retrieval and generation checks. The ideas graph projects structured evidence. It does not infer relationships directly from raw files.

Build a response in the Stoa

Let’s follow a question: “Seneca, I’m afraid of losing my job. How should I examine this fear?” The walkthrough illustrates the system’s operations. It does not start a real conversation.

Step 1 of 6
Your question follows these stepsYour messageQuestion addressed
Starting with Your message
  • @Seneca
  • I’m afraid of losing my job.
  • How should I examine this fear?
Select the participants
Result The requested voices
  • Seneca, explicitly addressed
  • Other guests where relevant
  • Prokopton42 for the synthesis
Illustrative walkthroughA selected voice may answer, ask for clarification or encounter a limit.

1. Decide who has something to contribute

Your message is read alongside the conversation so far and the thinkers invited to the room. Mentioning a thinker with @ requests their participation. Other invited thinkers may also speak if their texts have something relevant to offer.

Under the hood

The server checks conversation access and membership. Automatic routing assesses each invited voice’s relevance using retrieved evidence. An explicit mention requires selection, but does not exempt that voice’s response from validation. Accepted answers are passed to subsequent speakers so they can take account of what has just been said.

Your question follows these stepsWhat the question assumesContext explained
Starting with What the question assumes
  • A situation lived today
  • A problem to examine
  • Possibly unfamiliar words
Check the context needed
Result Before continuing
  • Enough context: proceed
  • Otherwise: explain the unfamiliar
  • Keep the explanation’s origin
Illustrative walkthroughExplaining a modern object does not give the thinker all the history since their death.

2. Separate their world from yours

Fear of losing your livelihood is a question an ancient thinker can discuss. Smartphones and social media may need an explanation first. The system looks for what falls outside the thinker’s period and considers the context you have already supplied.

Under the hood

Temporal checks compare the knowledge required with each person’s historical frame. Facts from ancient sources, the conversation, your documents and a modern explanation retain separate origins. When modern context is missing, the Quantum Leap setting chooses between a Temporal Guide intervention and asking you to explain.

Your question follows these stepsThe problem to look upPassage 13.4 proposed
Starting with The problem to look up
  • Fear of a future loss
  • Anxiety and anticipation
  • Sources admitted for Seneca
Retrieve, then check
Result A candidate passage
  • Moral Letters · 13.4
  • Fears that arrive before the facts
  • Text, translation and reference
Illustrative walkthroughThis is a possible passage, not a guaranteed result for every wording of the question.

3. Find texts that speak to this fear

Retrieval searches the sources permitted for the speaker. It might propose Seneca’s thirteenth letter, which examines anticipated fears. Translations of the same passage are brought together so they do not count as independent pieces of evidence.

Under the hood

The question is encoded in the same vector space as the segments. Retrieval returns candidates, then groups them by full passage identity. The primary text and provenance are resolved from the corpus. A relevance check excludes candidates insufficiently related to the question or its context. Attachments are searched separately within the authorised conversation.

Your question follows these stepsWhat the model receivesDraft composed
Starting with What the model receives
  • Your situation and the exchange
  • Selected passages and references
  • Voice and historical boundaries
Compose a response
Result A draft, still private
  • A response to your question
  • Proposed references
  • Quotations identified as quotations
Illustrative walkthroughNone of the draft’s wording is visible in the chat at this stage.

4. Write from the texts and your situation

The model receives your question, useful conversation context, selected passages and the voice’s rules. It proposes a response that applies this thought to your situation. A paraphrase is a new response; it must not be presented as an ancient quotation.

Under the hood

Ollama runs the model on the project’s infrastructure without calling a hosted model API. The request assembles role instructions, evidence and its origins. Structured JSON output separates the proposed text from references. For certain bounded topics, a previously reviewed formulation may be proposed; it passes the same checks before publication.

Your question follows these stepsThe draft under reviewDraft to check
Starting with The draft under review
  • The proposed response
  • Quotations and references
  • Attribution of information
Check for publication
AcceptedDisplay the response and its sources
Needs correctionLimit reached: stop and display a problem notice

Retry: the draft goes through the checks again

Illustrative walkthroughAn exact quotation can still be misinterpreted. Technical checks do not replace your reading.

5. Check before showing the response

The system checks references, words presented as quotations, information origins and specific historical rules. A rejected response may be corrected within a limited number of attempts. If it still fails, the chat displays a problem notice, not the rejected draft.

Under the hood

Validation checks include the output format, source identifiers, declared or visible quotations, length limits and rules about identity and chronology. Evidence excluded by the relevance check cannot be made citable again by a fallback. These checks are bounded: they do not demonstrate the soundness of every interpretation.

Your question follows these stepsThe accepted responsesResponse published
Starting with The accepted responses
  • Every voice stays attributed
  • Every citation opens its source
  • Disagreements remain visible
Compare the positions
Result The conversation continues
  • Prokopton42’s synthesis
  • Your reading of the passages
  • Your next question
Illustrative walkthroughThe optional reflection view is an educational reconstruction, not access to the model’s internal thoughts.

6. Let you read, check and respond

Each accepted response appears with sources you can inspect. After the historical voices, Prokopton42 offers a synthesis that distinguishes positions and their attributions. You can open a passage, challenge a reading and carry on the conversation.

Under the hood

The server sends progress to the browser through SSE, an event stream that avoids reloading the page. Text arrives as a complete response after validation, without streaming draft words. The synthesis may consult broader sources without attributing them to the thinkers. Depending on the room setting, further exchanges may follow within a fixed limit.

What if I ask about a smartphone?

When a modern concept needs explaining, there are two possible paths. Quantum Leap is the setting that allows the Temporal Guide to intervene.

“Seneca, why do I keep checking my smartphone?”

Quantum Leap enabled

The Guide supplies the context

  1. The need for a modern explanation is identified.

  2. The Temporal Guide intervenes for the relevant voices, explains the object, then leaves.

  3. The thinker examines your problem using this newly supplied information.

The explanation remains an outside contribution. It does not become a historical memory.

or

Quantum Leap disabled

The thinker asks you to explain

  1. No modern definition is added automatically.

  2. The thinker asks you to explain what they do not know.

  3. You describe the object; the discussion can continue with that context.

If your message already explains the object adequately, another question may be unnecessary.

Where does the information come from?

A response may draw on several kinds of information. Keeping them distinct prevents something you have just told the thinker from becoming an alleged ancient fact.

The thinker’s texts
They support historical claims within the limits of that voice’s admitted sources.
What you tell them
Your situation enters the exchange as your account, without becoming an established historical fact.
A document you attach
Its excerpts can help within this conversation. They remain material you supplied, separate from the historical corpus.
The Guide’s explanation
It supplies the necessary modern context and retains that outside origin.
Return to the passage in our exampleSeneca · Moral Letters · 13.4
Read the passage