Skip to main content

Agent Backend Tools

Tools are the execution interface between OMNI’s reasoning layer and actionable systems.

Tool Categories

Data Access

  • omni_api_* symbol/market/portfolio/model/factor/news/briefing tools
  • File text retrieval and artifact readers

Execution

  • python_run for computation and artifact generation
  • Plan submission tools for structured workflows

Utility + Validation

  • mermaid_validate
  • Prompt library search/get helpers

Registration Model

All built-ins are registered in backend tool registry and exposed based on active preset.

Preset Exposure

  • skills preset: product-facing safe tool surface
  • full preset: advanced developer surface (limited access)

Execution Flow

  1. Model selects tool call
  2. Tool schema validated
  3. Permission checks applied
  4. Tool executes with context
  5. Result streamed and persisted

Safety Constraints

  • Permission mode enforcement
  • Source-level allow/deny checks
  • Per-tool risk classification
  • Output guard and redaction controls

Design Tips For New Tools

  • Keep schemas strict and explicit
  • Return concise, parsable output
  • Avoid hidden side effects
  • Prefer first-party API adapters over direct external calls