Skip to content

How it works

From your sources to a cited answer.

Anuvrit keeps the original of what it reads, with its version. Answers are built only from evidence the asker is allowed to see, and every citation is checked before the answer is shown.

Capture keeps the original.

Code, tasks, documents and voice notes are read with their source and version attached. The archive stores each source once, compressed, and keeps every version by storing only what changed.

  • GitHub capture

    Pulls repository code, pull requests and review history into company memory.

  • ClickUp capture

    Pulls tasks, status and comment history from ClickUp.

  • Document and web capture

    Keeps documents and web pages with their sources and versions.

  • Voice note capture

    Turns voice notes into searchable text that keeps its source.

  • Exact deduplicated archive

    Stores each source once, byte for byte, so repeated content is not kept twice.

  • Lossless archive compression

    Compresses archived sources and rebuilds them exactly when read.

  • Version history by changes

    Keeps every version of a changed document by storing only what changed.

Fourteen steps from source to answer.

Ingest, authorise, retrieve, answer. A citation that does not hold up is rejected before anyone sees it.

Diagram: sources are captured, versioned, extracted, normalised, chunked, embedded and indexed. A query is authorised, candidates are retrieved and packed into bounded context, an answer is drafted, citations are validated and support is verified, and only then is the answer returned with citations. Citations that fail validation are rejected.

Answers you can check.

An answer is only useful if you can verify it. Each one carries its evidence, and each citation points to the exact source and version.

  • Access is checked twice

    Every passage is re-authorised before and after the answer is drafted. If access was revoked in between, the evidence is reported as unavailable.

  • Unsupported claims are dropped

    Claims the sources do not support are removed and the gap is named, instead of being filled with a guess.

  • Disagreement is shown

    When sources contradict each other or are too thin, the answer says so rather than picking a side.

  • Citations against exact bytes

    Search stays separate from the archive, and citations are checked against the archived original.

Give your agents your team’s history.

Anuvrit is in beta and opening to engineering teams in stages. Join the waitlist and we will contact you when there is a place for your team.