From birth data to a cited reading.

You give Iovari a date, time, and place. We resolve coordinates and IANA timezone, convert local time to UTC, and store the exact instant. Time accuracy is tracked explicitly — exact, approximate, rectified, or noon-default — so every downstream claim knows how much it can lean on the houses.

The natal chart is computed by the Swiss Ephemeris, the same engine that drives professional desktop tools (Solar Fire, Astrodienst). Positions, declinations, house cusps, aspects — the math is settled. Iovari never asks the LLM to "do astronomy"; the LLM only ever sees the structured output.

You pick the geometry: Tropical or Sidereal, Placidus or Whole Sign or any of seven house systems. The ephemeris version is recorded with every reading.

A strategy is a JSON document — bodies and points to include, aspect families and orbs, rulership scheme, dignities, sect, timing techniques (profections, zodiacal releasing, secondary progressions, transits…), and the narrative voice. Presets ship out of the box; Pro users build their own.

The recipe is the contract between you and the model. Whatever's in the recipe shapes the reading; whatever isn't, doesn't.

The LLM (Claude Sonnet by default; Opus on Pro deep-dives) sees the structured chart and the strategy. It has six tools: compute_chart, get_aspects, get_transits, find_returns, recall_memory, save_memory. It can ask for any calculation it needs but it cannot invent geometry — the math comes from the ephemeris, the interpretation comes from the model.

Every claim in the reading carries a citation tag pointing back to the placement / aspect / return / pin that produced it. The reading is stored with its (chart, strategy, model, prompt_version, ephemeris_version) tuple. Replay the same tuple, get the same reading. Drift is impossible.