Platform map
Find the five Noveum products, their shared data flow, and the separate developer and administration surfaces.
Noveum connects production observability with repeatable evaluation. Its five products cover observation, measurement, recommendations, synthetic testing, and runtime policy enforcement. Developer integrations, workspace administration, and reference material remain separate so they are not mistaken for additional products.
How the parts fit together
Each stage leaves an inspectable asset. That separation matters: you can change a mapper without losing raw traces, publish a new dataset version without overwriting an earlier one, or rerun scorers without generating new traffic. NovaPilot stops at recommendations; your team owns implementation, review, and deployment.
The five products
| Product | Main pages and assets |
|---|---|
| noveum-trace | Projects, organization Analytics, Explorer, project Traces, and the SDK integrations that create trace evidence |
| NovaEval | ETL jobs, datasets, eval jobs, scorers, Project → Runs conversation review, and Project Health |
| NovaPilot | Recommendations, report history, report detail, and scheduled reports |
| NovaSynth | Agent configuration, connections, personas, scenarios, synthetic runs, reports, and analytics |
| NovaGuard (beta) | Organization- and project-scoped cost caps and request or token rate limits for a connected runtime to enforce |
Start with The closed-loop workflow for the handoff between products and Choose the right dashboard for the actual organization and project routes.
Shared areas
| Area | What belongs there |
|---|---|
| Developer tools and integrations | REST and MCP references plus framework-specific guides; the Python SDK reference lives under noveum-trace |
| Administration | Workspace members, account security, billing and quotas, roles, and telemetry storage |
| Concepts and best practices | The tracing data model and guidance for designing traces, spans, attributes, and events |
Ways to connect
Application telemetry
The Python SDK provides context managers and integrations for LangChain, LangGraph, LiveKit, Pipecat, and CrewAI. It sends Noveum’s trace payload format using OpenTelemetry-aligned attribute names; it is not a general OTLP collector.
API and agent access
Organization API keys authenticate REST and MCP access. OAuth-capable MCP clients can instead connect through Noveum’s sign-in and consent flow.
API keys are organization-scoped. Do not put them in browser code, repositories, screenshots, or logs, and do not assume that a key for one organization can access another.
A practical first workflow
- Send a small number of development traces and inspect their structure.
- Add stable environment, service-version, session, and pseudonymous user attributes.
- Create an ETL job and validate its mapped items on a narrow trace sample.
- Publish a dataset version and run a few compatible scorers.
- Read failures at both the dataset-item and source-trace levels.
- Run NovaPilot only after the evaluation data is trustworthy.
- Have the responsible team implement one candidate change, then exercise it with NovaSynth before review and deployment.
- Add scheduled analysis only after the manual workflow produces decisions your team acts on.
The guided onboarding automates much of this first pass, while The closed-loop workflow explains the handoff between products.
Security and data handling
Trace payloads can contain prompts, responses, tool arguments, retrieved documents, transcripts, and audio. Allowlist or sanitize captured values in the application before sending customer traffic; optional SDK pseudonymization is only a second layer. Give members the narrowest practical role. For runtime cost and rate boundaries, configure project or organization NovaGuard policies, then verify that your gateway or runtime fetches and enforces the effective policy before provider calls and reports its decisions back to Noveum.
Support
- Dashboard: noveum.ai/app
- Documentation: noveum.ai/docs
- Email: support@noveum.ai