ChatAgents is an intelligent conversational assistant that breaks the knowledge-cutoff barrier by connecting your AI directly to the live web. Built on a hybrid Streamlit + FastAPI architecture, it uses LangGraph for agent orchestration and Tavily for real-time search, content extraction, and deep website crawling — so every answer can draw on the freshest information available, not just training data frozen in time.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/EllisYuan/ChatAgents/llms.txt
Use this file to discover all available pages before exploring further.
Architecture Overview
ChatAgents runs as two cooperating services that you can start independently for development or containerize together for production:| Service | Technology | Default Port |
|---|---|---|
| Frontend | Streamlit | 8501 |
| Backend API | FastAPI + LangGraph | 8080 |
Tech Stack
| Layer | Technology | Role |
|---|---|---|
| Frontend | Streamlit | Interactive chat UI |
| Backend | FastAPI | Async REST API + streaming |
| Agent | LangGraph | Agent graph orchestration |
| LLMs | Claude / OpenAI / Groq | Language model inference |
| Tools | Tavily | Web search, extract, crawl |
| Infra | Docker + Docker Compose | Containerised deployment |
Key Features
Real-time Web Search
Query the live web through Tavily’s AI-optimised search API. Fast mode returns 3 balanced results; Deep Thinking mode returns 5 with advanced depth.
Content Extraction
Precisely extract the key content from any web page URL, filtering noise and surfacing what the agent actually needs to answer your question.
Deep Website Crawl
Follow nested links to deeply crawl entire sites. Fast mode crawls up to 5 pages; Deep Thinking mode crawls up to 15, building a richer research picture.
Streaming Responses
Words appear token-by-token as the agent generates them. Tool calls — search, extract, crawl — are displayed inline so you can follow the agent’s reasoning live.
Session Management
Every conversation is assigned a unique session ID and persisted to disk. Switch between multiple sessions and revisit full conversation histories at any time.
Multi-Model LLM Support
Choose from Claude Haiku, Sonnet, or Opus for speed vs. quality trade-offs. OpenAI and Groq provider interfaces are available for teams that prefer those models.
Agent Modes
ChatAgents ships with two built-in operating modes that tune every Tavily parameter at once:- ⚡ Fast Mode — Uses
basicsearch depth, 3 results, a crawl limit of 5 pages, and no images. Best for quick factual questions where low latency matters. - 🧠 Deep Thinking Mode — Uses
advancedsearch depth, 5 results, a crawl limit of 15 pages, and image support. Best for in-depth research, competitive analysis, or multi-step reasoning tasks.
Prerequisites
Before you install ChatAgents, make sure you have the following:- Python 3.11 or later — required for the async features used throughout the codebase.
- Anthropic Claude API key — create one at console.anthropic.com. Keys start with
sk-ant-api-. - Tavily API key — register at tavily.com. Keys start with
tvly-. - OpenAI API key (optional) — needed only if you want to use OpenAI models instead of Claude.
- Groq API key (optional) — needed only if you want to use Groq-hosted models.
- Docker + Docker Compose (optional) — required only for the containerised deployment path.
What’s Next
Quickstart
Install ChatAgents locally, configure your API keys, and have a running chat agent with live web search in under 5 minutes.
Docker Deployment
Run both services as containers with a single
docker-compose up command — includes health checks and production Nginx guidance.