Deep framework integrations
LangChain, LangGraph, LiveKit, Pipecat, CrewAI, LlamaIndex, AutoGen, and OpenTelemetry: real instrumentation, not thin wrappers. The LangChain callback alone covers LLM, chain, agent, tool, and retrieval events.
TR
Observability
See every decision your agents make.
NovaTrace captures every LLM call, tool invocation, retrieval, and agent step: with tokens, cost, and latency attached. One SDK, deep framework integrations, and a transport layer that never takes your app down.
6M+
traces ingested per day
15,000
spans/sec, load-tested
15 min
from install to first trace
app.noveum.ai: traces explorer
01: Capabilities
Observability built for agents, not retrofitted.
Cost & tokens at capture time
An embedded registry spanning 40+ models across 11 providers prices every call as it happens: input, output, and total tokens plus cost on each span: so cost-per-trace and cost-per-customer are queryable from day one.
Threads & multi-agent graphs
Conversation threads, agent registries, and workflow graphs keep multi-agent systems legible: who did what, in which order, and why.
Multimodal artifacts
Audio and image artifacts upload alongside spans: the raw material that voice and vision evals need.
Fail-soft transport
Batched, retried, sampled, and silent on failure. Tracing never takes your application down: a guarantee, not a goal.
OpenTelemetry-aligned
Traces, spans, attributes, and events that map cleanly onto OTel concepts, with a purpose-built wire format for LLM workloads.
02: Integration
Live in fifteen minutes.
One init call plus the integration for your stack: and our engineers do it with you on the kickoff call.
pip install noveum-trace
from noveum_trace.integrations.langchain import ( NoveumTraceCallbackHandler,) handler = NoveumTraceCallbackHandler() agent.invoke( {"input": question}, config={"callbacks": [handler]},)What gets captured
- Spans for every LLM, tool, retrieval & agent step
- Token counts, model, cost, and latency per span
- Conversation threads and multi-agent graphs
- Audio & image artifacts linked to their spans
Python 3.9–3.13 · OpenTelemetry-aligned · MCP server included
03: Under the hood
The specifics behind the SDK.
A columnar trace store, open-source SDKs, an embedded cost registry, and an MCP server: the concrete pieces that make agent observability queryable.
Storage engine
A ClickHouse columnar backend purpose-built for trace volume: faster aggregations and materially lower storage cost than Elasticsearch on the same workload.
Open-source SDKs
Python `noveum-trace` (3.9–3.13) and TypeScript `@noveum/trace`: instrument with decorators and context managers, one init call to start streaming spans.
Model & cost registry
An embedded registry covering 40+ models across 11 providers: OpenAI (gpt-4o, o1, o3, gpt-5), Anthropic Claude 3/3.5/4.5, Google Gemini 1.5/2.0/2.5, Cohere, Mistral, Meta Llama, DeepSeek, xAI Grok, Perplexity, Amazon Bedrock, and AI21.
Per-span economics
Input, output, and total token counts plus computed cost are attached to every span as it’s captured: cost-per-trace and cost-per-customer are queryable from day one, no post-processing.
Native integrations
First-class instrumentation for LangChain, LangGraph, LiveKit, Pipecat, CrewAI, LlamaIndex, AutoGen, and OpenTelemetry: real callbacks, not thin wrappers.
MCP server
An MCP server with ~60 tools, generated from the OpenAPI spec, exposes traces, spans, and metrics to your own agents: query and act on observability data in-loop.
The loop
Part of one closed loop.
Everything downstream: evals, simulations, guardrails, fixes: runs on the traces NovaTrace captures. It’s step one for a reason.
Next step
Put your agents in the loop
Production AI agents your customers can trust.
Trusted in production by enterprise and high-growth teams