Astrum — Voice AI with Memory

REAL TALK FROM MISSION CONTROL

THE ARCHITECT & THE GRIND

Vitaliy Fedotov

It all started in December 2025 just for fun — wanted to mess around with voice AI and personal knowledge graphs. Thought I'd laugh, code a bit in the evenings. But the project sucked me in completely: I quit my regular job because I couldn't stop.

Now Astrum is no longer a hobby — it's a full-on thing: a hyper-advanced voice assistant running on your private knowledge graph, with 3D orbits, embeddings, spatial UI. I coded everything with Anthropic Claude — tried every coding agent out there (Gemini, Grok, o1, the rest), but only Claude has the memory and understanding to keep the whole complex project in its head without losing the thread. And yeah, it ate up every last dollar I had on API credits. Sitting in Calgary with zero savings, no investors, no corporate backing.

If the AI community and anyone who vibes with this doesn't throw some voluntary support my way — the project will just stall out. No moon promises, just a real indie dev trying to finish what he started.

WHAT'S AT STAKE

→ Mac Studio (M3 Ultra / 128 GB) — to run local models and stop bleeding money on Claude API
→ 5K monitor — my eyes are already dying from tiny pixels on code and orbits
→ Just to live — rent, food, coffee, so I don't have to freelance and drop Astrum

This isn't a hardcore crowdfunding with milestones. Just voluntary support from people who believe in private, powerful personal AI without Big Tech spying in your pocket.

SUPPORT ASTRUM 🚀 (KO-FI)

Every $10–20 = a few more hours of coding instead of survival mode. Thanks if you toss even a coffee.

LongMemEval Performance

TIMESTAMP: 2026-03-20 23:05:14 STATUS: IN THE PROCESS OF TESTING AND FINE-TUNING... LongMemEval Retrieval Results ══════════════════════════════════════════════ Overall Hit @AddressBook/Sources/Metadata: 373/500 (74.6%) By question type: knowledge-update 70/ 78 (89.7%) multi-session 70/133 (52.6%) single-session-assistant 51/ 56 (91.1%) single-session-preference 30/ 30 (100.0%) single-session-user 68/ 70 (97.1%) temporal-reasoning 84/133 (63.2%) Miss breakdown (127 total misses): sampling_cutoff 0 (0%) extraction_miss 8 (6%) ranking_miss 64 (50%) dataset_mismatch 55 (43%) Interpretation: ⚠ >33% misses from ranking — improve RRF scoring or temporal signals ══════════════════════════════════════════════ [LongMemEval] Results stored locally ✔ Test "Full oracle dataset" passed after 10061.366 seconds. ✔ Suite "LongMemEval" passed after 10061.367 seconds. ✔ Test run with 1 test in 1 suite passed after 10061.369 seconds.