sauravbhattacharya001/agentlens
AgentLens — Observability and Explainability for AI Agents
What's novel
AgentLens — Observability and Explainability for AI Agents
Code Analysis
10 files read · 3 roundsA self-hosted observability toolkit for AI agents that captures LLM calls, tool invocations, decisions, and token usage via a Python SDK, ships them to a Node.js/SQLite backend, and renders sessions as timelines, narratives, flamegraphs, and health scores through a no-build dashboard.
Strengths
Exceptional defensive programming (non-finite guards, buffer caps, secret redaction, path validation) combined with a clean separation-of-concerns architecture (types/render/orchestration split per feature) and thoughtful performance optimizations (single-pass aggregation, O(n) span tree building, reference-swap buffer drain). The SDK is genuinely production-ready with thread-safe transport, retry logic, and proper resource cleanup.
Weaknesses
The core tracing concept (instrument agent runs, visualize timelines) is not novel — it's essentially OpenTelemetry-style tracing applied to LLM agents, and the backend is a straightforward Express+SQLite CRUD store. The dashboard is a single 142KB JS file which, while functional, lacks the modularity of the rest of the project.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
205
Contributors
2
Files
238
Active weeks
5
Repository
Language
Python
Stars
1
Forks
0
License
MIT