Agents

pyCSAMT agents provide AI-assisted and rule-based workflow automation for survey loading, quality control, static-shift correction, phase analysis, inversion preparation, interpretation, reporting, and orchestration. Start with the orchestrating chat agent for guided end-to-end runs, browse the catalogue when you need one specialised agent, and use the family pages for the full reference of every agent group.

• 1. Workflow Orchestrator
  • 1.1. When To Use It
  • 1.2. Input Contract
  • 1.3. Routing And Reproducibility
  • 1.4. Output Contract
  • 1.5. Supported Workflow Families
  • 1.6. Running A Real Workflow
  • 1.7. LLM-Assisted Routing
  • 1.8. Inspecting Planned Steps
  • 1.9. Validation, Warnings, And Failure Modes
  • 1.10. Recommended Operating Pattern
  • 1.11. Related API
• 2. Assistant And RAG
  • 2.1. What The Assistant Solves
  • 2.2. High-Level Architecture
  • 2.3. Corpus Policy
  • 2.4. Chunk Types And Metadata
  • 2.5. Index Building And Persistence
  • 2.6. Current Corpus Statistics
  • 2.7. Retrieval Model
  • 2.8. Optional Dense Retrieval
  • 2.9. Context Assembly
  • 2.10. Generated-Code Validation
  • 2.11. Memory And Project Context
  • 2.12. Evaluation Suites
  • 2.13. Representative Retrieval Examples
  • 2.14. BM25 And Dense Retrieval Comparison
  • 2.15. Assistant Recipe Authoring And Review
  • 2.16. Direct Inspection Workflow
  • 2.17. Reproducibility Checklist
  • 2.18. Limitations
• 3. Agent Overview
  • 3.1. Core ideas
  • 3.2. How the pieces fit
  • 3.3. When to use agents
  • 3.4. Installation
  • 3.5. The AgentResult contract
  • 3.6. No-LLM request parsing
  • 3.7. Dry-run workflow preview
  • 3.8. Direct agent execution
  • 3.9. Coordinated workflows
  • 3.10. Natural-language orchestration
  • 3.11. LLM-assisted interpretation
  • 3.12. AI and model-zoo entry points
  • 3.13. CLI and web interface
  • 3.14. Outputs and reproducibility
  • 3.15. Where to go next
  • 3.16. Related API
• 4. Agent Catalogue
  • 4.6. How to read the catalogue
  • 4.7. Catalogue groups
  • 4.8. Choosing the right entry point
  • 4.9. Foundation and survey intake
  • 4.10. Processing and diagnostics
  • 4.11. Forward and inversion workflows
  • 4.12. AI and model-zoo agents
  • 4.13. Orchestration, pipeline, and outputs
  • 4.14. Typical chains
    • 4.14.1. Quality-control chain
    • 4.14.2. Correction and inversion-preparation chain
    • 4.14.3. AI inversion chain
    • 4.14.4. Batch workflow chain
  • 4.15. Minimal examples
  • 4.16. Support interfaces
  • 4.17. Related API
• 5. Agent And LLM Configuration
  • 5.1. Quick decision guide
  • 5.2. Configuration objects
  • 5.3. What gets configured
  • 5.4. Supported providers and defaults
  • 5.5. Recommended setup with environment variables
  • 5.6. Explicit Python setup
  • 5.7. How agents resolve LLM settings
  • 5.8. Per-agent overrides
  • 5.9. Multi-provider sessions
  • 5.10. Temporary overrides
  • 5.11. Running without an LLM
  • 5.12. Budget caps
  • 5.13. Pricing and rate overrides
  • 5.14. Inspecting active configuration
  • 5.15. Resetting configuration
  • 5.16. Practical recipes
    • 5.16.1. Offline QC recipe
    • 5.16.2. Local analyst recipe
    • 5.16.3. Mixed-model recipe
    • 5.16.4. Application session recipe
  • 5.17. Troubleshooting
  • 5.18. Configuration checklist
  • 5.19. Related API
• 6. Agent Coordinator
  • 6.1. What the coordinator solves
  • 6.2. Core objects
  • 6.3. Minimal workflow
  • 6.4. Step registration
  • 6.5. Input mapping with input_fn
  • 6.6. Dry-run preview
  • 6.7. Checkpoints and resume
  • 6.8. Required and optional steps
  • 6.9. Workflow result structure
  • 6.10. Complete QC and correction example
  • 6.11. Inversion preparation example
  • 6.12. Custom agents
  • 6.13. Best practices
  • 6.14. Common mistakes
  • 6.15. Related API