VARIA AI SUITE
Beyond chat windows. Varia is a local-first desktop agent suite that executes scripts, audits workspaces, creates assets, and self-corrects compile errors automatically under your control.
[ INFERENCE SWITCHBOARD ] INTERACTIVE PROVIDER SIMULATION
Configure the provider setting for chat, autopilot automation, and screen analysis in real time.
[ THE AGENTIC GAP ] WHY VARIA IS A SUITE, NOT A CHAT WINDOW
Raw LLM connections (like Gemini API or local Ollama alone) can write code but cannot test, execute, or correct it. Varia bridges this gap by introducing true system agency and local runtime loops.
| CAPABILITY | RAW CHAT CLIENTS (GEMINI / OLLAMA) | VARIA AGENTIC SUITE |
|---|---|---|
| SYSTEM AGENCY | None. Can only generate script text blocks for you to manually copy, paste, and run. | Autonomous Execution: Directly calls host OS commands, creates files, audits directory workspaces, and runs programs. |
| SELF-HEALING CODE | Zero compile awareness. If generated code crashes, you must copy/paste the error back to chat. | Auto-Repair Compiler: Automatically runs tests, captures error stack traces, finds failing lines, and writes self-patches until compile passes. |
| ASSET INTEGRATION | Text-only. Incapable of managing local generators or rendering audio layers. | Multi-Modal Pipelines: Integrates ComfyUI SDXL checkpoints and AudioMode engines, generating graphics and voice files locally. |
| MEMORY PERSISTENCE | Temporary context windows. History disappears or shifts out of window memory rapidly. | Qdrant Vector Database: Organizes persistent memory trees, logging context and roleplay facts across separate database indices. |
[ AUTOPILOT COMPILATION LOOP ] SELF-CORRECTING RUNTIME PIPELINE
Varia resolves errors autonomously before returning results. Observe the structural logic loop of Varia's Autopilot engine:
[User Request] ➔ [Mempalace Proxy (RAG Memory Audit)] ➔ [Smart Hybrid Inference Router]
│
▼ (Cloud Gemini / Local Ollama)
[Program Success] 🠜 [Compiler Pass] 🠜 [Quality Guardian Preflight Scan] 🠜 [OS Command Executor]
│ ▲
▼ │ (Fail: Parse Stack Trace & Edit Code Block)
[User Notification] [Self-Repair Hook]
compileall or unit tests). If compilation fails, the **Self-Repair Hook** intercepts the terminal traceback, highlights the exact line numbers, automatically executes a file modification patch, and re-runs the compiler check. The user is only notified once the program compiles clean.
[ CORE CAPABILITIES ] TECHNICAL SPECIFICATIONS & ARCHITECTURE
Mempalace Proxy Gateway
Centralized gateway for chat histories, tool registration, and vector retrieval. Coordinates local Qdrant memory collections to store context securely, enabling private RAG workflows.
Durable Speech FIFO
Audio Mode manages wake-word activation and local speech-to-text. Serialized playback is managed via a SQLite FIFO queue, preventing overlapping chatter and prioritizing notifications cleanly.
Autopilot OS Executor
Drives visual tools, script compilation, and email operations. Features a custom multiplayer TTRPG game-master mode that responds to email threads with automated public updates and private whispers.
Quality Guardian
A static/runtime checker integrated into the startup hook. Blocks launch if it detects hardcoded absolute directories, leaking API secrets in logs, or invalid local model registrations.
Factory Reset Baseline
Wipes tasks, schedules, memories, event streams, and vector collections with a single command, restoring Varia to a factory-fresh configuration while preserving your multi-gigabyte models.
Service Lifecycle Manager
A unified daemon orchestrator tracking exact PIDs for proxy, Ollama, Chronos, and ComfyUI servers, avoiding broad system process-name termination and ensuring clean, timed exits.
Truth Only / TTRPG Modes
A built-in system prompt trust engine. When roleplay_mode is disabled (default), Varia enforces a strict Truth Only Doctrine—preventing the AI from pretending to run commands or faking logs. Switch on TTRPG mode (/roleplay on) only when creative simulation is desired.
[ DEPLOYMENT MATRIX ] HARDWARE & SYSTEM FOOTPRINTS
| SPECIFICATION | MINIMUM (OFFLOADED / CPU) | RECOMMENDED (GPU RUNTIME) |
|---|---|---|
| SYSTEM RAM | 32 GB RAM (Requires Swap Support) | 64 GB RAM (Optimized Headroom) |
| GRAPHICS VRAM | 8 GB VRAM (RTX 3060/4060) | 24 GB VRAM (RTX 3090/4090) |
| STORAGE SPACE | 100 GB SSD (Model Files) | 250 GB NVMe M.2 SSD |
| SUPPORTED OS | Linux / Windows 10+ | Fedora, Ubuntu, Arch / Windows 11 |
[ LIABILITY WARDEN & LEGAL DISCLAIMER ]
IMPORTANT: Varia AI is a personal hobbyist project developed strictly for individual development workflows and localized research applications. It is not an enterprise product and has not undergone testing in diverse system configurations, security environments, or hardware baselines.
BY USING OR ACCESSING THIS DEMO REPOSITORY, SOFTWARE, OR DOCUMENTATION, YOU EXPLICITLY ACKNOWLEDGE AND AGREE TO THE FOLLOWING TERM OF LIABILITY:
