Docs disagree. Currant checks which won.

Currant finds which documentation claim was superseded, shows the model the winning evidence, and returns a receipt anyone can replay.

Try the live demo

Every page says it is current.
Currant checks which one is.

It reads the pages your agent retrieved, finds the claims that replaced older ones, and checks every supporting quote word for word before anything touches the context.

contested
quotecut
pageverifiedreceipt
replaced bypull
replay

How a question moves

Three hand-offs between a question and a checked answer.

One rule runs through all of it: nothing is cut, annotated or pulled unless our own code found a verbatim quote supporting it in the source text. The model proposes. Our code checks.

  1. Model, then our code
    01

    Find what replaced what

    A model reads the pages and proposes which claim superseded which. Every quote it cites is searched for, word for word, in the stored page. Quotes that don’t verify are thrown away.

  2. Our code
    02

    Put the winner in front of the model

    Spans are cut, not documents: a page that says “previously X, now Y” survives. When two pages disagree and nothing orders them, the answer is CONTESTED. Currant never guesses a winner.

  3. Our code
    03

    Return a receipt anyone can replay

    The receipt hashes the exact text the model saw. Replaying recomputes that hash, with no key and no network, and it matches or it fails.

Reasoning

SERV Reasoning, by OpenServ does every model call: proposing claims, proposing which replaced which, and answering, with one model, gpt-6-luna.

Judging

Reppo gave independent verdicts: its judges scored our answers against a baseline in 21 public evaluations.

How it works, in full

Live demo

Same pages. Two very different contexts.

Ask: “What are the API rate limit and the maximum upload size?” The pages are invented (docs.example.com) and the claims were written by hand so the demo is deterministic. It shows the mechanism, not how well the detector does on real documentation.

context.txt · the text the model is shown


        

The button recomputes the SHA-256 of the text above with your browser’s own crypto and compares it with the receipt. The command-line currant replay goes further: it rebuilds the context from the claim store and re-checks every quote.

Guarantees

What holds on every call, whatever the model says.

The model can miss a conflict. It cannot make Currant act on text that isn’t in the page. These rules are enforced in our own code, not left to the model.

  • Verbatim or nothing

    Nothing is cut, annotated or pulled unless our code found the supporting quote word for word in the stored page. The claim store refuses an unverified quote.

  • Spans, not pages

    Only the superseded sentence is cut. A page that says “previously 60, now 120” keeps its history.

  • No guessed winners

    When two pages disagree and nothing in them says which came first, both are marked CONTESTED, with both quotes.

  • Receipts that replay

    Each receipt holds the SHA-256 of the exact text the model saw. Replay recomputes it with no key and no network.

  • Fails closed

    If the proxy cannot ground a question, it answers 503 and sends nothing upstream. A question is never quietly answered ungrounded.

  • Crawled text is data

    Pages reach the model only behind SERV’s prompt guard and are never followed as instructions. The guard can fail open, so the checks that matter run in our code.

Plain retrieval, and the same pages through Currant
WhenPlain retrievalWith Currant
A sentence has been supersededIt stays in the contextIt is cut; the replacement and the line that proves it are pulled in
Two pages disagreeThe model picks one, silentlyBoth are marked CONTESTED, with both quotes
The model cites a sourceNothing checks itDropped unless the quote is found word for word
AfterwardsNo record of what the model sawA receipt that replays to the same hash

Real conflicts, checked by hand

Reppo’s own docs disagree with themselves.

Currant flagged the first two. We found the third by hand because the detector missed it. Every quote below is word for word from a saved copy of the page, checked by code in the repository. We reported all three to the maintainers.

Cost of a text-only evaluation

Flagged
docs.reppo.ai · staleAn evaluation costs 100 credits.
eval.reppo.xyz/agents.md · currentA text-only evaluation costs **440 credits**;

Measured: evaluations on 2026-11-05 and 2026-11-11 were each charged 440.

Sign-up bonus

Flagged
docs.reppo.ai · staleSign-up bonus | 1,000 credits (10 evaluations)
eval.reppo.xyz/agents.md · currentNew accounts start with **22,000 credits** (a one-time sign-up bonus).

The two pages cannot both be current.

Largest attachment

Missed
docs.reppo.ai · staleAttachments | 1 per evaluation, up to 2 MB: PNG, JPEG, WebP, or PDF
eval.reppo.xyz/agents.md · currentMax 1 item, 1.25 MB each

Found by hand. This is the detector’s recall limit on real documentation.

What we measured

The numbers, including where we lost.

The held-out figures come from 93 synthetic test items, scored once after a public freeze. The audit is Reppo’s own judges scoring our answers on real conflicts.

Precision0.857on 28 cuts, held out (dev was 1.000)
Recall0.82824 of 29 held out (dev was 0.892)
Stale page only84.2%answers right with Currant when only the stale page was retrieved, against 0% without
Reppo audit−7.7Currant against baseline on 3 real conflicts: 59.0 against 66.7. No conclusion from 3 questions
Reppo audit, 2026-11-12, one judgement per question
SetBaselineCurrantReading
15 easy numeric facts100100Identical answers: no harm, and no room to show a lift
Cost conflict8136We lost: “contested” named neither value
Sign-up bonus conflict5371Judges credited spotting the contradiction
Attachment conflict6670Detector missed it, so both gave the stale 2 MB

Why we lost points

  • When two pages disagree and nothing orders them, our answer says only that the documentation is contested. It does not quote either value.
  • The judges marked that incomplete and preferred a confident “440” that matched the live charge.
  • Fixed after the audit: in answer-v2 a contested answer lists each value the notice quotes, and our code drops any value the quotes don’t state. These scores are from v1 and stay as recorded; v2 has not been re-audited.

What this does not show

  • It does not show a lift on real documentation. The lift is measured only on synthetic items, with one model.
  • It does not show recall on real docs: the detector missed one of three conflicts, and we have no labels.
  • On a second vendor’s public docs (59 pages, 463 claims verified of 611 proposed) it ran with no code change and found nothing to report. That shows it runs elsewhere, not that it works there.

Integrate

Change one URL. Keep your agent.

currant serve is an OpenAI-compatible proxy. Point your client’s base_url at it: each question is grounded from the claim store first, then forwarded to SERV under your own key.

POST /v1/chat/completions
# 1. start the proxy (it listens on 127.0.0.1:8787)
uv run currant serve --extract-run 1 --relate-run 2

# 2. call it like any OpenAI-compatible endpoint
curl -i http://127.0.0.1:8787/v1/chat/completions \
  -H "Authorization: Bearer $SERV_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-6-luna",
       "messages": [{"role": "user",
                     "content": "What is the API rate limit?"}]}'

# the reply carries the receipt id
X-Currant-Receipt: <64 hex characters>
Holds no key
Your Authorization header goes to the upstream and nowhere else, and requests are not logged.
Fails closed
No store, no matching page or an unfinished run: a 503, and nothing is sent upstream.
No silent retry
A timeout is reported as unknown, with a 504. The caller decides whether to ask again.
A receipt per answer
Every grounded call writes a receipt and its exact context. The built-in viewer replays them.
Not supported yet
Streaming is refused with a 400. Only the last user message is grounded.

Questions

What people ask first.

More answers, and what is not built yet, are on the how it works page.

Why not just trust the newest page?

Pages rarely carry a reliable date, and a page that looks newer can still be the stale one: Reppo’s main docs site describes an API that its own agents.md has replaced. Currant only orders two claims when the pages themselves say which came first: a line such as “this replaces the previous value”, or a date printed on each page. Otherwise it says CONTESTED.

Why should I trust what the model proposes?

You don’t have to. Nothing is cut, annotated or pulled unless our own code found the supporting quote word for word in the stored page, and the claim store itself refuses an unverified quote. The model can miss things; it cannot make Currant act on text that isn’t there.

What happens when it cannot tell which claim is current?

It marks both as CONTESTED in the context, with both quotes, and never guesses a winner. Today the model’s answer then says the documentation is contested without naming either value, which Reppo’s judges marked as incomplete.

Does it delete whole pages?

No. It cuts the superseded sentence only. A page that says “previously 60, now 120” keeps its history.

Do I have to change my agent?

Only its base_url. currant serve speaks the OpenAI chat completions format and forwards to SERV under your own key. Streaming is not supported yet.

Where does my data go?

Replaying the demo is entirely offline. Running the pipeline sends pages and questions to SERV under your key. Nothing goes to Reppo unless you submit a file you reviewed, and Reppo evaluations are public.

Give your agent docs it can stand on.

Replay a receipt yourself, with no key and no network. The demo store is built from invented pages and hand-written claims, so a stranger can reproduce it end to end.

terminal
# no key, no network
uv sync
uv run currant replay examples/demo/receipt.json \
    --db examples/demo/store.db
# the quotes above are verbatim in the snapshots
uv run python scripts/build_drift_page.py --check