Open extraction protocol · bring your own key
Search the Pramana knowledge base
Add papers to the Pramana knowledge base
Pick a topic and this page runs the Pramana extraction pipeline on open papers, using your own model key. Two readers extract each paper independently. Fields they agree on are kept, each with a pointer to the sentence it came from, and you can publish the result as a bundle on Logos Storage.
The data layer holds only what each paper reported. Interpretations live in separate signed lenses. Every record carries a pin: SHA-256 hashes of the schema and prompts, plus the model IDs that produced it.
Keys
Choose a model provider and paste a key. Reader A and reader B should differ in model or temperature so their errors are less likely to coincide.
The page appends /chat/completions unless the URL already ends with it. The endpoint has to allow requests from a browser (CORS).
Your key goes from this page straight to the provider you chose, and nowhere else. There is no Pramana server in between. Unless you tick the box above, the key lives only in this tab's memory and is gone when you close it.
Topic
Type a topic or a claim you have seen. The protocol will look for open papers that measured it.
Design
The design model proposes search queries and topic-specific fields on top of the core claim fields. Review and edit them before anything runs. The pin below changes whenever you do.
Waiting for a topic.
Core fields follow the Pramana claim model: an assertion (who said it, the quote) kept apart from the proposition (subject, outcome, measure, relation, direction, magnitude, population, conditions, modality). Add or remove fields under topic_fields.
Pin
Recorded with every output so anyone can rerun the same protocol and compare.
Papers
Open-access works from OpenAlex and Europe PMC. Where Europe PMC holds the full text, the readers get Methods, Results and tables; otherwise they read the abstract, and the record says so.
Waiting for a design.
Extract twice
Reader A and reader B each read every selected paper on their own and return strict JSON with verbatim quotes. Neither sees the other's output.
Select at least one paper.
| Paper | Text | Reader A | Reader B |
|---|
Agreement
The two readings are compared field by field with fixed rules: normalized text match, 1% tolerance on numbers, and a check that every quote actually occurs in the text the readers saw. Disagreements go to a judge model or to you.
Waiting for extractions.
Publish
Each paper becomes a protocol record. The records and a manifest form a bundle, and the bundle's SHA-256 is part of its file name. Download it, add it to Logos Storage from your own node, and paste the content ID back here.
Waiting for agreement results.
Preview the first record
Publish to Logos Storage
A browser page cannot reach your Logos node, so run these in the folder where the bundle was saved. The last line prints the content ID (CID).
My contributions
Bundles you publish and record here are listed on this device only.
The reference knowledge base
The Pramana node has published a snapshot of 1,055 papers across seven topics: coffee and cardiovascular disease, diabetes, aging, caffeine mechanisms, repetitive negative thinking, AI and data center water, and urban simulation. Its content ID is
zDvZRwzm1sq9mZA6qVdigHnzPFos7zbzH4QEukk3VLMjhZTxDnGB
Anyone running a Logos node can fetch it by that ID. If a single byte differs, the ID does not match.
logosctl daemon start --detach logosctl call storage_module downloadToUrl "zDvZRwzm1sq9mZA6qVdigHnzPFos7zbzH4QEukk3VLMjhZTxDnGB" "$PWD/pramana_snapshot.json" false 65536 false false