TraceLLM
LLM observability and tracing platform for real AI applications
About TraceLLM
TraceLLM connects prompts, retrieval, tool calls, model spans, token usage, errors, and exports into an inspectable workflow. It wraps real model, Agent, RAG, and tool calls through an SDK without changing where requests run; sessions, spans, events, errors, usage, metadata, and optional content can be linked to one interaction, with control over content storage, redaction, sampling, and span types. TraceLLM supports OpenAI, Claude, Gemini, custom gateways, internal tools, and MCP-style workflows, and can forward selected traces to SigNoz, Honeycomb, Tempo, Datadog, or an OTLP HTTP collector.
Best for
Best use cases
Model call tracing
RAG retrieval and tool-call observability
Agent workflow debugging
AI application trace export
Features
Unified trace context
Links sessions, spans, events, errors, token usage, metadata, and optional content to one interaction.
Controllable data capture
Select content, metadata, errors, usage, redaction, sampling, and span types as needed.
Multi-provider and workflow support
Works with OpenAI, Claude, Gemini, custom gateways, internal tools, and MCP-style workflows.
External trace export
Forwards selected traces to SigNoz, Honeycomb, Tempo, Datadog, or an OTLP HTTP collector.
Pros
- Brings prompts, retrieval, tool calls, model spans, token usage, and errors into one workflow.
- Wraps models, Agents, RAG, and tool calls through an SDK without changing where requests run.
- Controls capture of content, metadata, errors, usage, redaction, sampling, and span types.
- Exports traces to SigNoz, Honeycomb, Tempo, Datadog, or an OTLP HTTP collector.