Deep Research
A multi-source agent swarm that researches like a graduate seminar — and shows its work.
Ask any serious theological or historical question. TheoSumma decomposes it, queries the licensed corpus, consults internal experts, watches the open web, builds a live knowledge graph, and writes a citation-grounded report. Export it as a PDF, present it as slides, or publish it to theosumma.com.
What Deep Research actually does
- Decomposes your question into sub-questions before spending a single token.
- Pulls from the TheoSumma library, consults named AI Experts, fetches Scripture, and queries the open web — then verifies every claim against the source.
- Builds a live Neo4j knowledge graph of every entity, claim, and citation it touches. You watch it grow in real time.
- Writes a research-paper-length report (≥ 2,500 words Standard, ≥ 5,000 words Exhaustive) and translates it into 14 languages with canonical-quote preservation.
- Composes a storytelling slide deck — hook, tension, evidence, resolution — with curated iconography. Present it in-browser or export as PDF.
How it differs from "ask an AI"
Chat answers something fast. Deep Research answers something well. The swarm runs 8–32 LLM-driven roles — Planner, Source-router, Retrievers, Entity-linker, Claim-extractor, Doctrine-checker, Bias-detector, Critic, Conflict-resolver, Synthesizer, Verifier — until the stopping policy says the question has been honestly handled. Every claim in the final report carries a citation back to its source. Nothing is fabricated; the Verifier catches it if the model tries.
Why this exists
Real theological questions don't get serious answers from a single LLM prompt. They take a graduate seminar's worth of decomposition, source triangulation, doctrinal cross-checking, and honest writing. Deep Research compresses that into ~3 minutes (Standard) or ~8 minutes (Exhaustive) — with the receipts to back the conclusions.
Frequently asked questions
What kinds of questions does this handle well?
Historical: "What did the Council of Trent decide about the canon of Scripture and how does it compare to the Reformed canonical position?" Doctrinal: "How do Catholic and Eastern Orthodox theologians differ on the procession of the Holy Spirit?" Exegetical: "What does the Greek of Philippians 2:6–11 actually say, and how have major traditions read the kenosis?" Apologetic: "What is the strongest historical case for the empty tomb?" Anything that benefits from multiple sources, careful citation, and honest weighing of competing positions.
Does the AI just make things up?
No. Every claim is grounded in a snippet the swarm actually retrieved. A dedicated Verifier role re-reads each claim against its source one at a time before the report finalises. Hallucinated quotations don't survive the Verifier pass.
What sources does it consult?
The admin-curated TheoSumma library (currently ~30,000 documents — Church Fathers, magisterial documents, councils, major theologians, biblical commentaries). Named AI Experts you can pick at task-start. The Bible in 14 translations. The public web through a privacy-preserving meta-search. Public YouTube transcripts. Public social posts. Everything you upload to the task.
Can I pick the denominational lens?
Yes. The Planner respects denominational scope per sub-question — you can ask the same question with a Catholic frame, an Eastern Orthodox frame, a Reformed frame, an Evangelical frame, or a neutral ecumenical one, and get appropriately different answers. "Where the traditions agree" and "where the traditions split" appear as honest sections in the report whenever the topic divides the Church.
Standard or Exhaustive — what is the difference?
Standard runs ~3 minutes, produces a focused ≥ 2,500-word report, and costs 1 from your monthly counter. Exhaustive runs ~8 minutes, produces a comprehensive ≥ 5,000-word report with broader cross-denominational coverage, and costs 4 from the same counter. Standard is enough for most questions; reach for Exhaustive when the question is contested or you need it citation-ready for academic use.
What's in the slide deck?
A storytelling agent reads the finished report and composes 8–12 slides on a strict narrative arc: hook, context, tension, evidence, resolution, invitation, closing. Each slide picks the layout that best tells its beat — icon hero, denominational comparison, real scripture card pulled from the bible service, pull-quote, timeline, stat, or gradient headline. Present it in-browser at fullscreen or download as PDF.
Can I publish my research publicly?
Yes — every finished report can be published at theosumma.com/research/<slug> with an SEO-friendly URL the AI generated from your question. A 24-hour editorial buffer applies before search engines index the page, giving the editorial review queue time to catch material that needs revisiting. Free-tier users help the platform by contributing their research publicly; paid tiers choose per-task.
How is this priced?
Each plan ships with a monthly Deep Research counter. Free includes 1 per month with mandatory public publishing. Standard 5, Plus 25, Pro 50. Exhaustive runs draw 4 from the same counter. Every 3 public publishes earns you 1 free Deep Research credit. No hidden charges, no per-token billing — you pick a plan and the counter resets monthly.
Can I refine a report after it's done?
Yes — up to 4 refinements per report. Refinement is a small targeted edit ("expand section 3", "add a Reformed perspective", "shorten the conclusion") that costs much less than a fresh research run. Refinements compose; the report keeps improving without losing its citations.
Does it work in my language?
The swarm thinks in English. A Translator role and a Language-refiner run at the tail to produce the final report in the language you picked — currently English, Arabic, German, Greek, Spanish, Persian, French, Hebrew, Hindi, Hungarian, Indonesian, Italian, Turkish, Urdu. Canonical Bible quotes are preserved verbatim from the published translation in your language.