Pegasus AI Platform with Voice, Vision and Agents
An AI assistant platform running entirely on local models. Supports image and file uploads, reasoning chat, real-time voice agents, live video interaction, and AI-powered web browsing where agents drive a browser to complete tasks. A node-based workflow system lets administrators swap or upgrade models without redeployment.
The challenge
The client wanted an AI assistant that could stand beside the hosted products, covering chat, reasoning, image and file understanding, voice and live video, but running entirely on local models so no data left their infrastructure, and without locking the platform to any single model.
What we did
Pegasus was built on FastAPI and Next.js with LangChain and LangGraph for orchestration and Ollama serving the models locally. It supports image and file uploads, reasoning chat, real-time voice agents, live video interaction, and AI-powered browsing in which agents drive a browser to complete tasks. A node-based workflow system lets administrators switch or upgrade models without a redeploy.
The outcome
A production platform with chat history, memory management and a password vault, running on local models end to end, with the model layer swappable from the admin interface as new open-weight models arrive.
By the numbers
Start your project
Ready to build something amazing? Let's discuss your project requirements.
What was built
Local model serving
Ollama-hosted open-weight models; nothing leaves the network.
Node-based model workflows
Administrators swap or upgrade models per task without touching code.
Voice and video agents
Real-time voice conversation and live video interaction.
Browser-driving agents
Agents control a browser to complete multi-step tasks on the web.
Vision and files
Upload images and documents; the assistant reasons over them.
Memory and history
Managed long-term memory alongside searchable chat history.
Technology stack
Technologies that powered this project.
Frontend
Backend
Database & Tools
Development process
Our proven methodology for delivering exceptional results.
Discovery & Planning
Understanding requirements, analyzing scope, and creating a detailed roadmap
1-2 weeks
Design & Architecture
System architecture, UI/UX designs, and technical specifications
2-3 weeks
Development & Testing
Agile development with continuous testing, code reviews, and QA
4-8 weeks
Deployment & Launch
Production deployment, performance optimization, and go-live support
1 week
More projects
Explore other successful projects in our portfolio.
RAG Chatbot over 100,000+ Posts
A retrieval-augmented chatbot embedded in a professional content platform, indexing over 100,000 posts in a Supabase vector database and answering strictly from stored content. An n8n workflow adds new posts to the index automatically, keeping the knowledge base current in real time.
Content platform
Publishing
AI Audio Annotation Platform
A platform accelerating audio data-labelling workflows: OpenAI models and PyAudio capture audio, generate transcriptions and help annotators produce structured annotations under strict validation rules. A companion browser extension auto-fills annotation fields on the DataForce TransPerfect site.
Data annotation client
Data Services
SmartQuery AI Agent
A full-stack AI chatbot in Django delivering context-aware conversations through LangChain, the Gemini API and RAG over a custom knowledge base. Includes session memory, Google and GitHub authentication, dynamic tool use, and integrated web, GitHub and LinkedIn search tools. Source is public.
Open-source project
Artificial Intelligence
Ready to build your next project?
Let's create something extraordinary together, with the same expertise and dedication shown in this project.
Build something like this
Service behind this project
Build something similar
More case studies
- RAG Chatbot over 100,000+ Posts — Case Study
AI project for Content platform
- AI Audio Annotation Platform — Case Study
AI project for Data annotation client
- SmartQuery AI Agent — Case Study
AI project for Open-source project
- View the Full Portfolio