Deneb

Personal AI gateway for NVIDIA DGX Spark. Telegram bot interface, Go gateway server, and Rust core engine — single-user, single-machine deployment.

Architecture

Telegram (Android) ──> Go Gateway (HTTP/WS) ──> Rust Core (FFI)
                           │                        │
                       130+ RPC methods         Protocol validation
                       Session management       SIMD memory search
                       Chat/LLM pipeline        Media processing
                       Telegram bot plugin      Context engine
                                                Vega semantic search
                                                GGUF inference (CUDA)

Module	Language	Description
gateway-go/	Go	HTTP/WS server, RPC dispatch, session management, chat/LLM pipeline, 130+ tool integrations, Telegram bot
core-rs/	Rust	Protocol validation, security, media processing, memory search (SIMD cosine + BM25 + FTS5), context engine, Vega semantic search, GGUF inference
proto/	Protobuf	Cross-language type definitions (Go + Rust codegen)
cli-rs/	Rust	CLI entry point, connects to gateway via WebSocket

Build order: Proto schemas -> Rust core (static lib) -> Go gateway (links Rust via CGo)

Prerequisites

• Rust (stable, via rustup)
• Go 1.24+
• buf (latest) + protoc + protoc-gen-go
• NVIDIA DGX Spark for GPU inference (optional — CPU fallback available)

Build

# Full build (Rust + Go + CLI)
make all

# Minimal (no GPU)
make rust && make go

# DGX Spark production (Vega + ML + CUDA)
make gateway-dgx

# Development (debug Rust + auto-restart Go)
make rust-debug && make go-dev

Test

# All tests (Rust + Go + CLI)
make test

# All checks (fmt + lint + test)
make check

Deploy

Single-machine deployment on DGX Spark:

git pull
make gateway-dgx
scripts/deploy.sh

Documentation

Full docs at docs.deneb.ai.

License

MIT