Obyflow
A CLI-first, local-first observability tool for LLM applications that traces and stores LLM calls, vector-store queries and LangChain steps in SQLite, then uses an LLM of your choice to help investigate the root cause of failures.
Obyflow traces and debugs LLM calls, vector-store queries and LangChain steps from the terminal, storing structured events, traces, logs, metrics, errors, tool calls, embeddings and custom events, locally in SQLite and connecting them into traces, with distributed tracing across services and automatic trace propagation.
It detects anomalies against historical baselines, diagnoses chain failures, tool timeouts, retriever issues and slow vector queries, tracks what changed across deployments, commits, configs, flags, models and dependencies, surfaces similar past incidents and their resolutions, and can run an AI investigation over the collected telemetry as evidence, choosing from providers such as Gemini, Anthropic or OpenAI. It also tracks token usage, cost and context limits, retries on rate limits and transient provider errors, redacts sensitive values before storage, and exports to JSON, CSV or OpenTelemetry OTLP.
Built with TypeScript and Python, with a CLI and Node.js and Python SDKs that integrate with LangChain, vector databases and embedding providers, installable through npm or pip.