Why Feenion Exists
AI engineering has exploded with abstractions—prompt chains, vector databases, rerankers, agents, and multi-tool workflows. Yet developers are often left guessing why an agent spent 8 seconds in a loop or why a retrieval step failed.
Existing SaaS observability platforms force developers to send their confidential customer prompts, proprietary business documents, and API keys to third-party cloud vendors. Feenion was created to provide a 100% self-hosted, open-source alternative where all telemetry stays under your complete control.
Core Design Principles
Zero cloud vendor lock-in. Run with Docker in one command on your laptop or private cloud.
Every screen answers questions: Is my AI healthy? What changed? Where is the bottleneck?
Decoupled asynchronous background workers ensure telemetry collection never slows user requests.
Licensed under Apache 2.0. Inspect, modify, and extend any part of the codebase.
Built by Darshan
Feenion was built by Darshan as an engineering project to explore distributed tracing in AI applications from first principles. I enjoy building developer tools that demystify complex software systems and give engineers full clarity over their applications.
Project Roadmap
- • Core Python SDK Tracing
- • Non-Blocking Async Exporter
- • Waterfall Distributed Timeline
- • D3 Mind Map DAG Visualizer
- • Error Fingerprinting Intelligence
- • Advanced Loop Detectors
- • RAG Quality Scorer
- • Trace Regression Diff Engine
- • Multi-Environment Filters
- • OpenTelemetry OTLP Ingest
- • Local Trace Replay CLI
- • Automated Evaluation Metrics