How filip-michalsky/salesgpt works
Project type and purpose
• From the README and file tree, this is a full-stack open-source AI Agent platform (“SalesGPT”) plus companion documentation sites.
• Core functionality: a Python library and backend API that orchestrates LLM-powered sales agents with tools (Stripe, Calendly, product catalog lookup, Mindware, etc.).
• Frontend: a Next.js React app for real-time chat.
• Two documentation sites:
• api-website (Sphinx-built library docs)
• website (Docusaurus marketing/site docs)
File structure
Top‐level directories & files:
• salesgpt/ ← core Python package (agents, chains, tools, prompts, logger)
• run.py ← CLI entrypoint for an interactive agent
• run_api.py ← backend API server (FastAPI/Flask)
• frontend/ ← Next.js React chat UI, Dockerfile.frontend, package.json, src/…
• api-website/ ← Sphinx docs for the Python library
• website/ ← Docusaurus marketing/documentation site
• examples/ ← sample configs, product catalog, notebooks
• tests/ ← pytest tests for library and API
• Dockerfile.backend, docker-compose.yml, Makefile, .github/… (CI), pyproject.toml, requirements*.txt
Architectural patterns & hints:
• Separation of concerns: UI vs. API vs. core library vs. docs
• Microservices‐style via Docker Compose: “backend” service, “frontend” service
• Python package is composable via run.py (CLI) or run_api.py (HTTP API)
• Tools are plugged into agents through LangChain custom agent pattern
README for additional insights
• Quickstart shows how SalesGPT.from_llm(...) wires up LLM + tools
• Architecture section includes a sample flow chart (agent → tool invocation → observation)
• External dependencies: LiteLLM (any LLM), Mindware integration, Stripe, Calendly, LangSmith tracing