Cogyto serves knowledge-driven organizations by empowering executives, support teams, and analysts with instant insights from documents and business data, enabling all employees to save time and focus on higher-value work.
- Smart Memory – Upload reports, meeting notes, or manuals; Cogyto remembers and retrieves details instantly.
- Real-Time Knowledge – Combines internal data with up-to-date web search for accurate answers.
- Proactive Assistant – Summarizes documents, drafts or sends emails, and automates routine tasks.
- Multilingual Support – Understands and responds in both English and Indonesian.
- RAG Workflow – Uses a plan → execute → reflect cognitive loop for accurate, refined answers.
- Backend: Flask, SQLAlchemy
- Database: PostgreSQL (pgvector)
- Knowledge Graph: Neo4j
- LLM: LLaMA
- APIs & Tools: Tavily, SerpAPI (search), Hugging Face Transformers (NLP/vision), smtplib (email)
- Frontend: HTML, CSS, JavaScript (real-time chat UI)
- LLM Hosting: LLaMA via Amazon Bedrock
- Vector Store: Amazon Aurora PostgreSQL (pgvector)
- Storage: Amazon S3 (documents, backups)
- Backend: Flask/FastAPI with SQLAlchemy
- Orchestration: LangChain / LlamaIndex
- Frontend: HTML, CSS, JavaScript (streaming-enabled)
cogyto/
├── run.py # App entry point
├── config.py # Config & env variables
├── app/
│ ├── routes/ # API endpoints (chat, kb, main)
│ ├── services/ # Business logic (chat, uploads, sessions)
│ ├── agents/ # Agentic framework (executor, prompts, memory manager)
│ ├── tools/ # Modular tools (web search, vision, email, knowledge graph)
│ ├── models/ # SQLAlchemy models (conversations, etc.)
│ └── utils/ # Text utils (cleaning, translation, language detection)
├── static/ # Frontend assets (HTML, CSS, JS)
└── docs/ # Documentation- Ask or Upload – Users start by typing a question or uploading documents.
- Plan – Cogyto,Creates a step-by-step reasoning plan to solve the query.
- Execute – The system calls the right tools (vector search, knowledge graph, web search, vision analysis, etc.) to gather information.
- Reflect & Refine – Reviews the draft answer, checks accuracy, and improves clarity before finalizing.
- Deliver – The user instantly receives a refined, contextual response enriched with insights.
- Add multimodal support (images, videos)
- Smarter long-term memory & personalization
- Caching & batch reranking for speed
- Improved UI with workflow visualizations