VOID

AI Incident Intelligence for Autonomous Agents

VOID is an open-source platform that transforms raw OpenTelemetry telemetry into actionable engineering incidents for AI agents.

It sits on top of observability platforms like SigNoz, helping engineering teams detect, prioritize, investigate, and resolve failures in complex agentic systems.

---

Architecture

Telemetry (SDK)
        │
        ▼
     SigNoz
        │
        ▼
 Ingestion Adapter
        │
        ▼
Incident Qualification Engine
        │
        ├── Deterministic Rules
        ├── Fingerprinting
        ├── Sampling
        └── Severity
        │
        ▼
LLM Evaluator
        │
        ▼
Replay Generator
        │
        ▼
Issue Agent
        │
        ▼
Repair Suggestions

---

Repository Structure

void-ai/
│
├── .github/
├── void
├── void-sdk
├── void-server
├── void-dashboard
├── void-demo
├── void-examples
└── void-docs

Repositories

Repository	Purpose
void	Architecture, roadmap, documentation, screenshots and demos
void-sdk	OpenTelemetry SDK for AI agents
void-server	Incident intelligence engine
void-dashboard	Web dashboard
void-demo	Demo agents with reproducible failures
void-examples	Framework integration examples
void-docs	Research and design documentation

---

What VOID Does

VOID converts telemetry into engineering intelligence.

Instead of only showing traces, it helps answer:

• Why did the agent fail?
• Which traces belong to the same incident?
• Is this failure new or recurring?
• How severe is the incident?
• What should engineers fix first?
• How can the incident be reproduced?
• Can a repair suggestion be generated?

---

Components

VOID SDK

Responsible for instrumenting AI applications.

Agent
    │
    ▼
Tool
    │
    ▼
Custom Span
    │
    ▼
OpenTelemetry
    │
    ▼
SigNoz

---

VOID Server

Core intelligence layer.

api/
ingestion/
risk-engine/
evaluator/
fingerprint/
incident/
replay/
issue-agent/
repair/
scheduler/
storage/
prompts/
models/
telemetry/
utils/

---

Risk Engine

rules/
detectors/
samplers/
policies/
classifier/
signals/

Example detectors

• LoopDetector
• RetryDetector
• TokenExplosion
• ToolMismatch
• ContextOverflow
• TimeoutDetector

---

Evaluator

hallucination/
tool-selection/
reasoning/
confidence/

Responsible for LLM-based incident analysis.

---

Fingerprinting

hash/
clustering/
dedup/
similarity/

Example

Incident-145

contains

38 traces

---

Replay

trace-reconstructor/
timeline/
prompt-view/
tool-view/
memory-view/

Replay an AI execution from telemetry.

---

Issue Agent

summarizer/
github/
report/
templates/

Outputs

• Engineering Report
• GitHub Issue
• Markdown Summary

---

Repair

cursor/
codex/
patch/
verification/

Generates repair suggestions while remaining isolated from the rest of the system.

---

Dashboard

Dashboard
Incidents
Replay
Reports
Rules
Settings

The dashboard focuses on engineering workflows rather than raw observability.

---

Demo Applications

weather-agent/
customer-support/
rag-agent/
broken-agent/

Example failures

• Tool mismatch
• Infinite loop
• Timeout
• Hallucination
• Empty retrieval

---

Examples

langgraph/
crewai/
openai/
google-adk/
pydantic-ai/

---

Documentation

architecture/
design/
adr/
images/
research/
pitch/

---

Infrastructure

collector/
signoz/
postgres/
redis/
docker/
kubernetes/

---

API

/traces
/incidents
/rules
/replay
/reports
/repair
/health

---

Fixtures

good-trace.json
loop.json
tool-mismatch.json
timeout.json
hallucination.json
context-overflow.json

Used for deterministic development and testing.

---

Schemas

Trace.ts
Incident.ts
Risk.ts
Fingerprint.ts
Replay.ts

Shared contracts across services.

---

Roadmap

Current MVP

• SDK instrumentation
• Incident qualification
• Fingerprinting
• LLM evaluation
• Replay generation
• GitHub issue generation

Future

• Automated repair suggestions
• Multi-project support
• Production deployment
• Kubernetes support
• Advanced analytics

---

License

MIT