Compile a document set into a reviewable work output

What this is for

Take a small document set through compile, review, activation and one question, and end with an answer you can open at the source location it came from.

Before you start

Sources: PDF, DOCX, XLSX, PPTX, ODT, ODS, ODP, JPG, PNG, TIF or GIF. Every one of them is sanitized to PDF and read by OCR before anything else happens.

Tables and formulas are not extracted as tables: a figure arrives as the paragraph it was printed in, with the page and the box it sat in, so a number is readable and the grid that arranged it is not. Heading levels, sections and columns are not recovered as typed structure; what the compiler receives is a paragraph and where it sat. TAVONEL processes sources up to 5 MB and 80 pages.

A compile produces a candidate, and a candidate becomes an active World only when a person activates it in the workspace. That step needs a paid plan: the Developer plan, if you are the workspace owner, or the Team plan, which is sold through a conversation today rather than self-serve checkout. A free evaluation can upload, compile, review and export; its activation request is refused, so decide the plan before you plan the run.

Fix the questions before the run rather than after it: one plain fact, one that needs two documents together, one conditional comparison, and one the sources cannot answer. The last one is the useful one.

What a free evaluation includes, and what a page costs, are on the pricing page. This page does not restate either, so it cannot disagree with them.

Running it in code

The runnable path -- upload capability, PUT the bytes, start a compile, poll the job -- is published in the API quickstart with cURL, Python and TypeScript, and it stops where the product stops: a key cannot activate a World.

The exact contract: Quickstart in the documentation.

Where this stops

The result is a draft for a person to check. This workflow is not used to settle a medical, legal or investment question, and a document set is only as current as its last compile.

The reading limits under "Before you start" apply to this recipe too.

Next action

Open the read-only sample to see a compiled World and its evidence without an account, read the quickstart if you are starting from a key, or talk to us about activating a World on your own corpus. Nothing on this page sits behind an email form.

If you would rather not run it yourself, the World Build is a fixed-scope engagement on your own corpus. The scope and what it leaves behind are below; pricing is set with you — talk to us.

The structure: customer provides representative corpus, TAVONEL compiles, evidence / gaps / review states shown, current World, one update/change test, grounded Ask, architecture + economics review. What it leaves behind: the compiled World itself as a reviewable candidate, a source and evidence report, a report of what could not be compiled and why, one update demonstration, a latency and cost snapshot for that corpus, and an implementation plan for going beyond the pilot.

Two things are settled before any document moves: every format in the corpus has to be one TAVONEL accepts, and the corpus may not need the customer-data gate that is off by default.

Run and revision information

Every section above describes what the product does today, and the revision of this guide is printed at the foot of the page. A verified run date appears there only once a person has recorded a run, never on an edit to this page.

What a recorded run will add

The actual input — The document set the run used, published from a corpus cleared for redistribution.

Following it on screen — A capture of the workspace screens as the run passed through them, recorded rather than staged.

What the run produced — The files, tables and drafts the run produced, as it produced them.

Checking it against the source — The source location behind each value, opened at the page and the region it came from.

Using the result outside TAVONEL — The result an external consumer read back from the same World, verified on its own run.

Measured effect — Quality, time, cost and human effort, from a same-condition measurement of that run.