SignalMesh: A Multi-Agent Fault Diagnosis System Based on LangGraph

SignalMesh is an open-source multi-agent workflow system for O&M incident triage developed by the developer maharanasunil1843, built on LangGraph. It adopts a dual-agent architecture (analyst + report agent) to automate the processing from raw telemetry data to structured fault reports, addressing the pain points of traditional manual troubleshooting such as time-consuming and error-prone. Core designs include type contract enforcement, conditional routing retry, and fail-safe mechanisms, providing O&M teams with a scalable and auditable automated diagnosis framework.

Background and Problems

In modern distributed systems, O&M teams face the challenge of massive monitoring data and alarm information. Traditional fault troubleshooting relies on manual analysis of logs, metrics, and tracing data, which is not only time-consuming and labor-intensive but also prone to missing key information or making misjudgments. With the increasing complexity of systems, automated and intelligent fault diagnosis has become an urgent need in the O&M field.

Core Architecture Design

SignalMesh adopts a dual-agent collaboration model, decoupled via type contracts:

1. Analyst Agent: The core reasoning engine, which calls telemetry tools to obtain data, analyzes root causes, and outputs type-safe structured findings (AnalystFinding).
2. Report Agent: Receives the output from the analyst and converts it into the final report, with no access to raw data to ensure consistency and auditability. In addition, the system has a built-in conditional router to implement bounded retry logic: if the confidence level is low, it retries at most once; if the retry fails, it enters the fail-safe node to generate an "unresolved" report, avoiding crashes or fabricating results.

Technical Implementation Highlights

1. Type Contract Enforcement: Define data structures via handoff_contract.py to ensure consistent formatting, verifiable interfaces, and runtime type checks.
2. Structured Observability: Each step generates structured logs, facilitating debugging, performance analysis, and optimization.
3. Task Success Measurement: Built-in functions quantify diagnostic effectiveness to support continuous improvement.
4. Offline Reproducibility: Uses a simulation provider by default, allowing operation without API keys and reproducible results, which is convenient for development testing and CI/CD integration; switching to a real model only requires configuring the .env file.

Use Cases and Value

SignalMesh provides O&M teams with:

• Rapid Fault Response: Automatically converts telemetry data into structured reports.
• Knowledge Precipitation: Encodes diagnostic logic through typed finding objects.
• Human-Machine Collaboration: Marks "unresolved" when the root cause cannot be determined to avoid misleading.
• Enhanced Observability: A complete trace link helps understand the diagnostic process.

Summary and Outlook

SignalMesh demonstrates the application potential of multi-agent systems in the O&M automation field. By enforcing type contracts, conditional routing, and fail-safe design, it solves common reliability issues of agent systems. Its architectural ideas (agent decoupling, bounded retry, honest failure) have important reference value for building production-level agent systems, and it is an open-source project worth in-depth study for engineers exploring AI-driven O&M solutions.