Learn the workspace. Check our work.
Guides for getting started, the methods behind every number, where the data comes from, and what changed in each release.
Your first thesis in ten minutes
Paste your notes, turn them into assumptions, and set your first goal.
Read the guide → GuideGoals and rules for the delegate
What it may do alone, what it asks about, and what it never does.
Read the guide → GuideAsking good questions for boards
How to phrase a question so the board answers it, and how to pin and rebuild.
Read the guide →Models propose. Code calculates.
Every figure in Annula is computed by deterministic code from a named source. Language models read, summarize and draft, and every step they take is logged with its inputs. They never produce a number you see.
- Discount to fair value is (fair value − price) ÷ fair value. Upside is (fair value − price) ÷ price. The labels never swap.
- Fair value history only shows values you adopted, with dates. A model run is never shown as your view.
- Missing data shows as a gap, not as a guess.
fees FY24 = $4,800M · 10-K p.33
(5,200 − 4,800) ÷ 4,800 = +8.3%
computed by: calculator · AI involved: none
Where the numbers come from.
Filings, financial statements, insiders, ownership, news and macro data, collected from the source, cleaned and cited. The full list, how fresh each source is, and the rules our collectors follow are on the Data page.
What's new.
Boards
Ask a question and the delegate builds a board of calculated, sourced blocks. Pin boards to the sidebar. Follow-ups rebuild them.
News mood and market periods
The price chart shades up and down periods and shows a news-mood strip. Click a period to see what drove it.
Goals and rules
The delegate now works inside goals, with rules for what it may do alone, what it asks about, and what it never does. Everything it does lands in Activity with an undo.
Filings reader
Read any filing in place, with passages matched to your assumptions and your notes in the margin.
Why we're building Annula.
Most research doesn't fail because it was wrong. It fails because nobody checks it again. A thesis written with care in January is a stranger by September, buried under filings and headlines nobody matched against it.
We think software can carry that load: keep the reasons attached to the position, do the reading and the math, and bring back only what needs a person. That's the whole idea.
The Annula team