How azzzwhy/data-analysis works
This is a local-data statistical analysis agent: a user asks a question, the configured model selects a tool, local code computes over the dataset, and the model explains the returned results. The main flow is supported by the README and sampled agent/skill excerpts. The skill also documents resident sessions, analysis planning, batch work, and report deliverables; links for unsampled subsystems are limited to what the tree and documentation support.
Main entry points: agent_main.py Tool-calling agent · gpu_analytics.py Statistical engine · skills.py Skill dispatch · gpu_session.py Resident sessions
Components 17
User experience
- Terminal UI
agent/tui_app.py - CLI entry
agent/agent_main.py
Agent and connections
- Connection setup
agent/api_setup.py - API configuration
agent/api_config.py - Tool-calling agent
agent/agent_main.py - Skill dispatch
agent/skills.py
Local analytics
- Statistical engine
skills/cudf-analytics/scripts/gpu_analytics.py - Resident sessions
skills/cudf-analytics/scripts/gpu_session.py - Analysis planning
skills/cudf-analytics/scripts/analysis_plan.py - Parquet cache
skills/cudf-analytics/scripts/parquet_cache.py - Report outputs
skills/cudf-analytics/scripts/make_deliverables.py
Extensions
- External tool policy
agent/external_tools.py
Other components
- Analyst
- Model provider
- Local dataset
- MCP servers
- Installed skills
Connections 24
- Analyst → Terminal UI: asks question
- Terminal UI → Tool-calling agent: runs query
- Analyst → CLI entry: asks question
- CLI entry → Tool-calling agent: starts agent
- Connection setup → Model provider: lists models
- Connection setup → API configuration: saves settings
- Tool-calling agent → API configuration: loads settings
- Tool-calling agent → Model provider: sends prompts
- Model provider → Tool-calling agent: returns tool calls
- Tool-calling agent → Skill dispatch: dispatches tools
- Skill dispatch → Statistical engine: runs analysis
- Statistical engine → Local dataset: reads data
- Statistical engine → Tool-calling agent: returns statistics
- Tool-calling agent → Model provider: sends tool results
- Model provider → Tool-calling agent: returns answer
- Skill dispatch → Resident sessions: runs session tools
- Resident sessions → Local dataset: loads data
- Resident sessions → Analysis planning: builds plan
- Resident sessions → Statistical engine: runs analysis
- Statistical engine → Parquet cache: uses cache
- Tool-calling agent → External tool policy: dispatches external tools
- External tool policy → MCP servers: calls approved tools
- External tool policy → Installed skills: discovers skills
- Skill dispatch → Report outputs: creates deliverables