Skip to content

Aware

Aware is a minimal Assess prototype for telemetry-based root-cause analysis. It combines a Parser agent, an Executor agent, and dynamic analysis sub-agents to turn an incident question into a validated BuildSpec and an execution report.

The primary interface is the aware CLI. A lightweight web UI is also available for live inspection of agent events and structured outputs.

The project focuses on a clear runtime flow:

  • parse the user request into a BuildSpec
  • select the relevant telemetry files
  • run specialized log, trace, and metrics agents
  • filter analysis on failure_time_range_ts
  • persist run artifacts as JSON, text, and optional SQLite records

Technology Stack

Python Google ADK OpenAI FastAPI Pydantic SQLite

Interfaces

Interface Use it for
CLI Running parser-only, executor-only, or full assessment workflows from a terminal.
Web UI Watching live agent events and reviewing structured outputs interactively.

Core Components

Component Role
ParserAgent Converts a user incident request into a validated BuildSpec.
ExecutorAgent Instantiates analysis sub-agents and coordinates execution.
tools/ Provides file loading, time-window filtering, and telemetry helpers.
templates/ Defines dynamic agent templates for logs, traces, and metrics.
Knowledge DB Stores runs, events, task results, and findings in SQLite when memory is enabled.

Next Steps