Sign in Join

Live chat for Free Pascal, Lazarus, Delphi and every Pascal programmer.

Catalog · How the catalog works
Anything · Lazarus or Delphi
Random projectRandom project

SaathiAI

View on GitHub

Member rating

No ratings yet.

Sign in to rate this project.

README

SAATHI AI 2.0 — AGENTIC DESKTOP OPERATING SYSTEM

Engineered by Pratham Prasad

Local-First Sovereign AI Co-Pilot • 60FPS Arc Reactor HUD • Multi-Agent Orchestration • 539-Chunk Local RAG Knowledge Engine

🌟 Overview

Saathi AI 2.0 is an omniscient, local-first agentic cognitive co-pilot and desktop operating interface. It combines a 60FPS animated 4th Civilization Arc Reactor canvas with multi-agent goal decomposition, local RAG vector indexing, zero-API-key inference, and sovereign Windows OS control.

Saathi AI 2.0 unifies 8 next-generation AI subsystems into a single cohesive platform while preserving 100% backward compatibility and zero external API key requirements.

🎨 Visualizer & UI Experience (4th Civilization Arc Reactor)

pascal
       .---.           [ 72 Micro-Degree Ticks ]
     /  ___  \         [ 3 Sentient Rings: Astral Violet, Quantum Teal, Solar Gold ]
    |  ( @ )  |        [ 96-Tooth Stator Turbine Gears ]
     \  ---  /         [ 12 Plasma Lavender Conduits + Micro-Lightning ]
       `---'           [ 48-Channel Audio Spectrum Equalizer ]
  • 60FPS Animated HUD: Real-time vector rendering engine with 3 concentric rotating energy rings, Fibonacci spiral vortex arms, and state-adaptive plasma lightning (THINKING, PLANNING, EXECUTING, SPEAKING).
  • Radar Diagnostic Beam: Holographic 360-degree sweep line displaying system vitals, core temperature, and memory stream hex telemetry.
  • 48-Channel Equalizer: Live spectrum analyzer responding dynamically to voice perception and Edge TTS speech synthesis.
  • Real-Time Task Observer: Dedicated side drawer displaying live step-by-step sub-agent execution graphs.

🚀 Integrated Subsystems & Architecture

Saathi AI 2.0 consolidates eight state-of-the-art AI frameworks into a unified tool ecosystem:

Subsystem Module Path Capabilities & Description
WebCmd saathi/tools/web_cmd.py @agentrhq/webcmd integration for Playwright browser sessions, live web page extraction, and web command execution.
SelfEdifyAI saathi/tools/self_edify.py Recursive self-improvement, continuous performance reflection, memory synthesis, and self-optimizing prompt pipeline.
Acontext saathi/tools/acontext.py Dynamic token context compression, multi-turn state preservation, and memory vector indexing.
Claw Code saathi/tools/claw_code.py Fast Rust-accelerated code parser, AST security sandbox verification, and multi-file code refactoring.
Ponytail saathi/tools/ponytail.py Autonomous multi-agent task orchestrator, goal breakdown graphs, and tool routing pipeline.
DeepSeek Harness saathi/tools/deepseek_harness.py DeepSeek Harness (dsh) plugin architecture (AGENTS.md), reasoning mode optimization, and local LLM task routing.
PrevJarvis saathi/tools/prev_jarvis.py Legacy Jarvis desktop core, Tauri app bridge, Rust acceleration backends, and desktop UI control.
Graphify saathi/tools/graphify.py Tree-sitter AST codebase knowledge graphs, community detection, and force-directed graph querying.

🧠 Knowledge RAG & Multi-Agent Engine

  1. SQLite RAG Knowledge Store (data/knowledge.db):
    • 539 stored knowledge chunks covering all integrated subsystems, documentation, architecture specs, and benchmarks.
    • Full-Text Search (FTS) and vector retrieval for sub-millisecond query responses.
  2. Master Multi-Agent Orchestrator (SaathiOrchestrator):
    • Automatically deconstructs complex goals into a multi-agent task graph.
    • Dispatches tasks across CodingAgent, WebAgent, VisionAgent, AutomationAgent, MemoryAgent, ConversationAgent, and 100+ dynamic factory sub-agents.
    • Headless HTTP synthesis with auto-failover (Pollinations AI, OpenRouter Free, and local Ollama saathi-distill-brain:latest).
  3. 8,192 Token Prediction Budget:
    • Generous token prediction limits ensure large multi-line prompts and complex code generation never get cut off.

⚡ Quick Start & Usage

1. Installation & Environment Setup

Run setup.bat to initialize directory structures, install dependencies, and run system diagnostics:

powershell
.\setup.bat

2. Launch Saathi AI 2.0 Control Console

Launch saathi.bat for the interactive 10-directive management menu:

powershell
.\saathi.bat
pascal
===================================================================
              SAATHI AI 2.0 ??? MASTER CONTROL CONSOLE
                   Engineered by Pratham Prasad
===================================================================
 [1] Launch Saathi AI 2.0 HUD & Cognitive Interface
 [2] Run System Doctor Diagnostics (py -3.12 -m saathi doctor)
 [3] Run Modular Unit Test Suite (27/27 Verification)
 [4] Launch Continuous 24/7 Self-Advancement Daemon
 [5] Launch Autonomous Dev-Loop Engine (Overnight Self-Dev)
 [6] Push Repository to GitHub (Auto-Repair & Sync)
 [7] Build Android APK Package Configurations
 [8] Verify & Install Python Dependencies
 [9] Run System Integrity & Module Compilation Check
 [10] Exit Console
===================================================================

3. Launch Desktop GUI Directly

powershell
.\Start-Saathi.bat

🧪 System Diagnostics & Unit Tests

  • Run System Doctor Diagnostics:
    powershell
    py -3.12 -m saathi doctor
  • Run Full Unit Test Suite (27/27 Passing):
    powershell
    py -3.12 -m unittest discover -s tests -p "test_*.py"
  • Push Changes to GitHub:
    powershell
    .\push-to-github.bat "your commit message"

📄 License & Attribution

Designed and engineered by Pratham Prasad (2026).

Licensed under the Apache License, Version 2.0 and MIT License. You may obtain a copy of the License at LICENSE.

All integrated open-source submodules (WebCmd, SelfEdifyAI, Acontext, Claw Code, Ponytail, DeepSeek Harness, PrevJarvis, Graphify-8) are attributed to their respective open-source creators and unified into Saathi AI 2.0 under compatible open-source terms.