How devgomesai/ollama-cloud-crewai-research-assistant works
Project Type and Purpose
• This is a “full-stack” AI research assistant with a primary CLI interface and an optional HTTP API layer (“FastAPI”) for future extensions.
• It orchestrates a multi-agent workflow (Research Agent → Reporting Agent) via CrewAI, executing local LLM calls through Ollama and auxiliary web-scraping tools, then outputs timestamped Markdown reports.
Main Components
a. User Interfaces
• CLI (Rich-powered terminal UI)
• Optional HTTP API (FastAPI endpoint in src/api/api.py)
b. Orchestrator
• CrewAI engine (ResearchCrew in src/agent/src/agent/crew.py)
c. Agents
• Research Agent: gathers data using search/browsing tools
• Reporting Analyst Agent: synthesizes findings into markdown
d. Tools Layer
• DuckDuckGo Search, Website Browser (requests + BeautifulSoup), Current Time tool (src/agent/src/agent/tools/tools.py)
e. LLM Service
• Local Ollama server (configured by BASE_URL in .env)
f. Configuration
• YAML files in src/agent/src/agent/config/agents.yaml & tasks.yaml
• Pydantic settings (env vars in .env)
g. Persistence & Output
• Markdown reports in output/ (timestamp-named files)
• HTML template placeholder in src/templates/index.html
Relationships & Data Flow
• User → (CLI or HTTP API) → CrewAI Orchestrator → Agents (in sequence) → Tools Layer & Ollama → Agent outputs → Orchestrator → “output/” storage + CLI/API response
• Config files feed into Orchestrator at startup (agent roles, tasks)
Architectural Patterns & Principles
• Modular multi-agent pipeline (separation of concerns)
• Dependency injection of tools & LLM endpoint via configuration
• Clear separation between UI, orchestration logic, agent implementations, tools, external integrations, and data persistence
Technologies & Frameworks to Annotate
• Python 3.13 runtime
• CrewAI (agent orchestration)
• Ollama (local LLM execution)
• Rich (CLI UI)
• FastAPI (HTTP API)
• Pydantic (config management)
• Requests + BeautifulSoup (web scraping)
• LangChain (LLM utilities)
Elements
• Clear labels on all components
• Notes on config inputs (YAML, .env) feeding into Orchestrator