38+ tools, one loop
read_profile, update_profile, add_school, run_essay_eval, query_peers, save_scene, and more. The model picks; the loop dispatches.
§ Suite · Services
What Illuma actually does.
▸Live previewClick to collapse
Add MIT to my list as an EA target, and pull 5 peers who got in with a similar CS profile.
Done — added MIT to your list as an EA target (deadline Nov 1). Pulling peer matches now.
add_school→ MIT · EA · deadline 2026-11-01
query_peers→ 5 CS admits, 3.9+ UW · SAT 1510+
Admit3.98 UW · 1560
USACO Platinum, 2 papers
Hook · First-gen
Admit3.92 UW · 1520
Founded CS club (60 kids)
Hook · URM · TX
Admit4.00 UW · 1580
Regeneron STS finalist
Admit3.94 UW · 1540
MIT PRIMES cohort
Common thread — every one of these had one deep systems / research project. Your USACO Silver + mentorship arc reads well on the mentorship side, but you're light on the research/publication axis compared to these admits.
Output includes
- Opens with your profile + activities + honors + essays + college list already loaded
- 38+ callable tools: update_profile, add_school, run_essay_eval, query_peers, and more
- Writes hit the same DB your modules use — no forked state
- Peer search + admit-probability calls happen inline during chat
- Threaded history per student; counselors switch students without losing context
Under the hood
ML models + calibration layers, not a ChatGPT wrapper.
Same DB as the modules
When the chat updates your GPA, the Foundation module sees it. No sync job, no eventual consistency — one Postgres row.
Per-student threads
Counselors can pin a chat to a specific roster student. The agent reads that student's data, not the counselor's own.