From an idea to
an architecture.
SYRIX began as a voice assistant experiment and evolved into a guarded, multi-node AI runtime — version by version, module by module.
A proprietary, multi-modal An runtime
designed for high-performance automation.
Avananka was conceived by Kabirr Kohli in London, UK as a deeply personal project — an An that doesn't just respond to commands but lives on your machine. The name itself is an acronym that maps each letter to a technical pillar and a family member, embedding human meaning into every layer of the architecture.
Built on a precision-tuned rig — RTX 4060 Ti, i7-12700K, 32GB DDR5 — Avananka is engineered for 24/7 local-first runtime, provider-aware inference, and local memory storage.
| Pillar | Family | Architectural Role | |
|---|---|---|---|
| A | Autonomous | Amit | High-level logic & decision-making kernels |
| V | Virtual | Vaneeta | interface management & virtual environments |
| A | Assistant | Ahana | Primary interaction & UX/Un node |
| N | Network | Namita | External data streams & APn connectivity |
| A | Analysis | Amit | Data processing & market intelligence loops |
| N | Neural | Naveen | An brain, LLM routing & neural kinetics |
| K | Kinetics | Kabirr | Master Architect & physical system control |
| A | Automation | Ahana | Task execution & automated workflows |
| N | Node | Naresh | Connectivity, hardware sync & network security |
The evolution.
The birth of Avananka. A single-file voice assistant built on raw PyAudio and Piper/Edge fallback TTS. The foundation that every future node would build upon — a continuous while True loop listening for your voice.
The utility expansion. Memory was born — jarvis_memory.json gave Avananka a persistent brain between sessions. Live data feeds from Open-Meteo and RSS/free-web providers connected it to the world. Natural language reminders and audio control arrived.
Intelligence arrived. Ollama brought local LLM fallback onto the GPU, Groq Cloud added a fast provider route, and the newer router now supports NIM specialist roles. Screen OCR gave Avananka opt-in visual context, while dynamic audio ducking made the experience smoother.
The great refactor. Avananka became a true architecture — split into intent modules for Media, System, Web, Trading, and Utilities. Parallel threading runs wake-word and clap detection simultaneously. Proactive Vision watches your screen. This is the version that ships.
The connectivity era arrives. Avananka reaches beyond the machine — Telegram and WhatsApp integration brings it to your phone. Shazam Mode identifies tracks in real-time via shazamio. NZXT RGB sync creates ambient hardware feedback that mirrors system state.
Full local Llama 3 70B inference running entirely on the 4060 Ti. Complete offline mode — no internet dependency for any core function.
Predictive tasking, gesture control, and advanced neural routing. The system begins to anticipate before you speak.
Encrypted family nodes and full network hardening. Avananka becomes secure at every endpoint.
"The routed HUD-first runtime." Every node unified, every sensor opt-in, every high-risk action approval-aware.
The mark, refined.
Every version of the Avananka logo represents a chapter in the project's growth — each iteration sharper, more intentional.
