System Protocols

Cognito is a functional, tactical workstation for sovereign intelligence. This documentation outlines the exact technical standards for local deployment.

Architecture Overview

Designed for absolute locality, the stack is partitioned into three functional layers:

• Command Shell (Electron/React): A tactical UI built for high-legibility and local resource binding.
• Inference Core (FastAPI): The localized gateway service that maps neural instructions to the llama.cpp engine.
• Context Module (RAG): An in-memory retrieval engine using TF-IDF protocols to inject private knowledge archives into active context.

Boot Sequence

Node deployment requires an isolated environment. We explicitly support source-level builds to ensure operational transparency.

Prerequisites

• Node.js v18.0.0+ (Interface Runtime)
• Python v3.10.0+ (Processing Kernel)
• C++ Compiler (Hardware Layer Mapping: Metal/CUDA)

# CLONE_SOURCE
git clone https://github.com/ArjunDeshwal/cognitoai.git

# INITIALIZE_VENV
python -m venv venv
source venv/bin/activate
pip install -r backend/requirements.txt

# EXECUTE_INIT
cd app && npm install && npm run electron:dev

Retrieval RAG

Local knowledge injection is handled via the Context Module. Artifacts (PDF/TXT) are parsed and tokenized into 400-word segments.

Similarity is ranked via Cosine Weighting against the query vector, enabling the model to retrieve specific excerpts without external indexing services.

Agent Sensory

The system possesses Autonomous Discovery capabilities. When the neural core identifies an information void, it initiates a [SEARCH:] protocol to scan global networks.

Command API

Technical interfaces for direct node control:

Neural Mounting

POST/v1/load_model
Mounts a GGUF neural archive to the inference core.

Discovery Flow

POST/v1/chat/completions
Initiates a neural reasoning stream with agentic flags enabled.