dialogtr — 2026
DialogTR
An Android AI assistant where one team agent commands a fleet of tools.
ACT I — THE PROBLEM
A single monolithic prompt can't run a real assistant — search, scraping, weather, maps, email and a database all need different hands, permissions and failure modes.
ACT II — THE SYSTEM
A universal team agent built on Agno delegates to eight specialized tools behind a provider-agnostic LLM layer (Gemini, OpenAI, Claude — hot-swappable), with per-user tool permissions, MCP support, session memory and on-device Whisper for Turkish speech-to-text.
ACT III — THE NUMBERS
8
TOOLS ORCHESTRATED
3
LLM PROVIDERS
on-device
SPEECH-TO-TEXT
A working multi-agent assistant on Android: fully async FastAPI backend in a hexagonal architecture, JWT auth with blacklisting and rate limiting. Backend and mobile repos are public.
NEXT FILM
SponsorHub