Trace the ṭuruq
Gather multiple routes before reaching a conclusion.
Built for your most complex hadith research
Hadith Agent
Not a chatbot guessing at hadith. An agent for advanced hadith research — planning the workflow, tracing the sources, examining the evidence step by step, checking the chains, and grounding every answer in the classical corpus.
See it in action
A real question, start to finish.
The exact Hadith Agent interface — a question, a cited answer, an open source, a narrator profile.
The architecture
Retrieval first. Reasoning second.
The agent never answers from memory. It routes the question, gathers the real pages, composes their evidence, reasons across them, and traces every claim back to its source before a word reaches you.
-
01 Route
Know where to look before moving.
The question crosses a single gate. The agent reads its intent, then wakes only the relevant regions of the corpus before choosing a route through them.
- Intent classified
- Corpus territory selected
-
02 Retrieve
Visit the pages, not a memory of them.
The agent moves source to source across the corpus. Every page it opens is an actual piece of evidence, and each one stays on the record for the work that follows.
- 6–8 source nodes visited
- Evidence remains traceable
-
03 Compose
Bring the evidence into one cited file.
Fine threads are drawn from every gathered page toward one working centre. Nothing is flattened yet; each source keeps its own identity and route back to the corpus.
- Evidence assembled
- Source identity preserved
-
04 Reason
Connect evidence to evidence.
The threads begin to cross: route against route, narrator against critic, wording against commentary. Only after those relationships hold does the network contract into a conclusion.
- Relationships compared
- Conclusion formed downstream
-
05 Verify
Send every claim back to its source.
A verification pulse travels from the conclusion to each page that supports it. The answer is not released until every visible claim can complete that journey back.
- Claims traced one by one
- Citations checked before release
-
Return
A cited answer crosses the same gate.
The answer reaches you; the agent stays with the corpus. If the sources cannot support a claim, that claim does not cross the boundary.
The corpus is not a backdrop. It is the limit.
Your research method
Set the depth. Shape the researcher.
Choose how far the agent should go and how it should communicate the result.
Research settings
Grounded answer
A focused response for quick checks and source discovery.
Fast · concise · citedStructured research
Multiple searches, comparisons, and a deeper evidence trail.
Multi-step · comparative · citedMaximum depth
A long-form investigation for demanding research questions.
Extended · methodical · transparent
Expertise
General reader, student, or scholar
Focus & tone
Disciplines, detail, and custom instructions
Skills & Protocols
Ten disciplines. One method.
A plain language model answers from whatever it happened to read — no method behind the verdict, no chain of custody, no idea which narrator to trust. These ten skills are exactly what it lacks: each one a recognised protocol from the classical critics, injected into the agent as standing discipline. They are what make it mature in the real intricacy of ʿilm al-ḥadīth — a trained muḥaddith, not a fluent guesser.
Custom skills
Teach it your own protocol.
The built-in skills are a starting bench, not a ceiling. From the app's Skills & Protocols panel you can author your own — name a method, write its instructions, and the agent adopts it as standing discipline in every session. The more precisely you encode your school's method, the more mature and discerning its research becomes.
Openable evidence
Every citation opens into its source.
Citations are part of the research interface—not a footnote added at the end. Open a citation beside the answer, move between sources, copy it, bookmark it, or continue reading in Corpus Workspace.
- Inline references tied to gathered evidence
- Book, volume, and page context
- One-click source reader beside the conversation
- Unsupported citation tags removed before delivery
6 / 147 · No. 4748
العلل الواردة في الأحاديث النبوية
وسُئِل عن حديث الوليد بن كثير، عن محمد بن جعفر بن الزبير، عن عبيد الله بن عبد الله بن عمر، عن أبيه، عن النبي ﷺ في الماء وما يَنجُس منه.
فقال: يرويه الوليد بن كثير واختُلف عنه؛ فرواه بعضُهم هكذا، وخالفهم آخرون في شيخه وفي إسناده، والوجوه متقاربة المخرج.
1 / 2
Sanad-aware research
Research the report as a transmission, not just text.
Hadith Agent can trace routes, inspect narrators, and surface points that require ʿilal analysis—then take you into the dedicated workspace when the graph matters.
Inspect the narrators
Open source-backed profiles and critic quotations.
Surface the disagreement
Identify where routes diverge and which evidence needs study.
Research that accumulates
The next question starts where the last one ended.
A study you finish is recorded — the narrator, the verdict, the critics relied on. Ask about the same subject weeks later, in either app, and it is there. Nothing is re-derived from nothing.
Bound to the source
Every claim carries a locator you can open — never a loose quotation.
Across sessions
Findings live on the server, so they follow you between Claude and ChatGPT.
Confirmed by others
What a second researcher independently reaches is promoted, and offered to the next.
Corrections stick
Correct it once and the correction is kept — you do not fight the same error twice.
Initial inquiry
2 turns synced
It grows with you
It starts sharp. It grows wiser with you.
The more you use it, the more it matures — growing sharper, wiser, and more attuned to your research with every session you share.
Attunement to your research
deepens with every session
-
Early
Grounded, cited answers on careful house defaults.
-
Ongoing
It learns how you work and carries your corrections forward.
-
Seasoned
It reasons in your manhaj and picks up where you left off.
What it has learned about your research
Tied to your account
- Preference Answers in Malay, with the full isnād in every citation.
- Correction You grade Ibn Lahīʿah's late riwāya as ḍaʿīf — it stops leaning on it.
- Skill You lead rijāl with Ibn Ḥajar, then al-Dhahabī — it keeps that order.
- Research Active thread — ḥadīth al-tasmiya, picked up where you left off.
Everything it learns is your private memory — visible in Manage Memory, and yours to edit or erase anytime.
Pricing
One method. Choose which library it reads.
Everything above runs on every plan, the free one included. A plan decides which library becomes unlimited — not how good the method is.
$0
Both libraries, a few studies a day
The whole method, with a limited number of studies a day.
- Reads our hosted corpus
- And your own library, if you have one
- The limit is how many studies, never how deep
$4.58/mo
$55 billed yearly
Your shelves, read in place
Reads the Arabic library already installed on your computer. Nothing is uploaded.
- Unlimited studies
- Thousands of kitab you already own
- Books never leave your machine
- One command to install
$7.33/mo
$88 billed yearly
Our corpus, nothing to install
140+ curated kitab, hosted and indexed, with the isnad graph behind them.
- Unlimited studies
- Isnad graph and narrator index
- No install — add one connector
FAQ
Questions, answered.
Is Hadith Agent just another chatbot?
No. It's an evidence-driven research agent bound to a real classical corpus — it answers only from retrieved, cited passages, so every claim is traceable to a named source rather than generated from memory.
Where do its answers come from?
From a classical hadith library — the Kutub al-Sittah and many classical works. The agent retrieves passages from this corpus and cites each claim to its book, volume, and page, with a jump straight to the original text.
What kinds of research can it do?
Takhrīj, isnād (sanad) analysis, rijāl and al-jarḥ wa al-taʿdīl, authenticity grading, comparison (muqāranah), and in-depth study (dirāsah). You can also browse the classical books directly in the Corpus Workspace.
Does it remember my research?
Yes. It keeps a private memory tied to your account — your preferences, your corrections, and the threads you're working on — and grows more attuned to how you research the more you use it. It never overrides the sources, and everything it keeps is yours to view or erase anytime in Manage Memory.
What platforms is it on?
It runs in your browser today, with a Windows desktop app available now and a macOS app coming soon.
How do the Free, Pro, and Max plans differ?
All three tiers run the same 19 research skills in the same enforced order — the free tier is not a weaker method, only a limited number of studies a day. Local Library removes that limit and works over the Arabic library already on your computer. Corpus Workspace removes it too and works over our hosted corpus, with the isnad graph and narrator index behind it.
Hadith Agent