Local-First AI Data Analysis System
DataCamp built a comprehensive local data analyst application that runs entirely on-device using OpenClaw with Ollama for complete data privacy. The system accepts CSV datasets and optional supporting documents (PDF, TXT, Markdown), then performs multi-step analysis workflows to generate insights and visualizations without any external API calls.
Architecture and Workflow Implementation
The system operates through three core components: (1) Web interface (web_assistant.py) handling file uploads and creating run directories, (2) OpenClaw agent loading workspace skills and coordinating local tool execution, and (3) Analysis engine (main.py) reading datasets, inferring relevant columns, and generating charts. The workflow produces three key artifacts: trend_chart.png for visualizations, analysis_report.md for insights, and tool_trace.json for execution tracking.
Technical Setup and Configuration
Implementation requires OpenClaw installation (curl -fsSL https://openclaw.ai/install.sh | bash), Ollama setup with qwen3:8b model (brew install ollama; ollama pull qwen3:8b), and local configuration through .openclaw-local/openclaw.json pointing to the local Ollama instance at http://127.0.0.1:11434/v1. The system uses Docker container isolation and runs OpenClaw as non-root user for security.
Privacy and Performance Benefits
This architecture provides complete data privacy as all processing occurs locally, with no data transmitted to external services. The system works offline once models are downloaded, offers full transparency with inspectable execution traces, and delivers agent-style multi-step workflows triggered by single user requests. This approach is especially valuable for sensitive financial reports, legal documents, or any scenario where cloud AI solutions are unacceptable due to privacy requirements.