Python and TypeScript SDKs
- Logger refactor: Refactored the SDK logger in both the Python and TypeScript SDKs and added support for mandatory notices.
Python SDK v1.2.3b1
- OpenAI Codex SDK support: Added tracing support for the OpenAI Codex SDK.
This is a beta release. Install it explicitly with
pip install netra-sdk==1.2.3b1.Python SDK v1.2.2
- Gated baggage propagation: Added gated baggage propagation across HTTP instrumentations.
EvaluationsVoiceNew featureImprovementBug fix
Human annotation, AI test case generation, and new evaluation flow
Evaluations
- Human annotation: You can now add human annotations to single-turn and multi-turn test runs.
- New evaluation creation flow: Released a redesigned flow for creating evaluations, available in the US region.
- AI test case generation: You can now generate test cases with AI when creating a text evaluation. See Generate test cases with AI.
Voice
- Bug fixes: Fixed several issues in voice evaluations.
Python SDKVoiceSimulationsNew featureImprovementBug fix
Python SDK v1.1.0 and WebSocket voice agent testing
Voice simulations
- WebSocket voice agent testing: Automated testing of voice agents over WebSockets is now available in the US region. See WebSockets.
Python SDK v1.1.0
- Header redaction: Added the
NETRA_REDACT_HEADERSenvironment variable to configure which HTTP headers are redacted. - Lower memory use: Reduced memory use for non-streamed HTTPX and Requests responses.
- Less overhead when tracing is disabled: Input and output capture is now skipped on non-recording spans.
- Generator handling: Early generator closure is now treated as normal completion rather than a span error.
Alerts
- View Traces button: Alert notifications now include a View Traces button that opens the related traces in Netra. See Alert rules.
Python SDKTypeScript SDKSimulationsNew featureImprovement
Python SDK v0.1.98 and TypeScript SDK v1.8.0
Python SDK v0.1.98
- Hermes Agent tracing: Improved tracing and span hierarchy for Hermes Agent, and deduplicated tool execution spans.
- OpenAI improvements: Enhanced OpenAI span previews and streaming for the Responses API.
- Simulation hooks: Added prescript and postscript hooks for simulation execution. See the Python simulation reference.
TypeScript SDK v1.8.0
- Postscript failure status: Simulation hooks now report a failure status when a postscript fails. See the TypeScript simulation reference.
ObservabilitySimulationsTypeScript SDKNew featureImprovement
Trace resilience and TypeScript SDK v1.7.0
Platform
- Metadata truncation: Added truncation for large metadata values.
- Trace recovery: Added a trace recovery pipeline to improve trace resilience.
- Hook details in test runs: Prescript and postscript details now appear in the conversation preview within test run details.
TypeScript SDK v1.7.0
- Simulation lifecycle hooks: Added simulation lifecycle hook support for multi-turn evaluations. See the TypeScript simulation reference.
TypeScript SDK v1.6.0
- Root instrumentation configuration: Added configuration inside root instrumentation spans.
- Span attribute truncation: Enhanced truncation of span attributes.
Red TeamingPlatformObservabilityEvaluationsNew featureImprovement
Red Teaming launch, Get Started and trace view improvements
Red Teaming
- Red Teaming: Introduced Red Teaming to test your AI applications against adversarial attacks.
Platform
- Get Started: Renamed the Home page to Get Started.
Observability
- Error highlighting: Spans with errors or exceptions are now highlighted when you first open a trace.
- Faster observations: Optimized the Observations API query for faster responses.
Evaluations
- Rule-based online evaluators: Expected values for rule-based online evaluators are now entered manually.
TypeScript SDK v1.5.0
- Google GenAI instrumentation: Added instrumentation for Google GenAI. See Gemini.
- OpenAI Agents cache tokens: OpenAI Agents instrumentation now captures cache token details.
- Span input and output utilities: Fixed the utilities that set input and output on active and root spans.
TypeScript SDK v1.3.0
- OpenAI Agents native tracing: Added configuration for native tracing in OpenAI Agents.
Netra MCP
- Evaluations through MCP: You can now run evaluations through Netra MCP, including creating datasets and dataset items from a QA specification you provide.
Datasets
- List datasets by tag: Added tag-based dataset listing for SDK usage. See the Python evaluation reference.
TypeScript SDK v1.2.0
- Simulation file handling: Added file handling support in simulation workflows. See the TypeScript simulation reference.
AlertsDashboardsEvaluationsObservabilityNew featureImprovementBug fix
Alerts, dashboards, and Evaluator Playground updates
Alerts and dashboards
- Metadata filters in alerts: Added metadata filters to alert rules.
- Dashboard filters: Fixed filters in project-level custom dashboards.
- Widget preview range: Widget previews now default to a 24-hour time range.
Evaluator Playground
- JSON field values: Added support for JSON field values in validation and execution.
- Validation messages: Updated validation toast messages.
Observability
- Claude Agent costs: Fixed a cost mismatch in Claude Agent spans.
Python SDK v0.1.93
- Span attribute truncation: Fixed configurable truncation for span attributes.
SimulationsEvaluationsInsightsNew feature
Stress testing, simulation file upload, and multimodal evaluators
Simulations
- Stress testing: Released stress testing.
- File upload: You can now upload files in simulations.
Evaluators
- Multimodal evaluators: Rule-based and custom LLM evaluators are now available for multimodal use cases. See Image evaluators.
Insights
- Project-level configuration: Added project-level configuration for Insights.
TypeScript SDK
- OpenAI Agents SDK instrumentation: Added instrumentation for the OpenAI Agents SDK.
- HTTP instrumentation: Added HTTP instrumentation.
- Span blocking: Added a mechanism to block specific spans.
- Root and default instrumentation: Added root and default instrumentation configuration. See Instrumentation selection.
Evaluations
- Multi-image evaluation: Released multi-image evaluation. See Image evaluations.
Datasets
- Bulk edit: Added bulk editing for single-turn datasets.
Python SDK v0.1.84
- Instrument configuration: Added the
instrumentsandroot_instrumentsoptions. The SDK now enables a curated default set of instrumentations instead of every available one. See Instrumentation selection. - Claude Agent TTFT: Added TTFT and relative TTFT for Claude Agent instrumentation.
- Instrumentation fixes: Fixed the Agno and Claude Agent instrumentations.
Python SDK v0.1.83
- Agno instrumentation: Added instrumentation for the Agno framework.
- Span input and output utilities: Added utilities to set input and output on active and root spans.
- Trace internal class methods: Decorators can now trace internal class methods.
Python SDK v0.1.82
- ADK traces: Cleaned up the Google ADK trace hierarchy and fixed related bugs.
- Span input and output: Added support for input and output span attributes.
Python SDK v0.1.81
- Relative TTFT: New LLM spans now record relative Time to First Token.
- HTTP payload capture: Added custom HTTPX and Requests instrumentation that captures request payloads and response details in span attributes.
Insights
- Insights (beta): Released Insights in beta to surface user intents and drift in your production traffic.
Evaluations
- Online Evaluation (beta): Released Online Evaluation in beta to score production traces automatically.
Datasets
- CSV upload: You can now upload CSV files to create datasets anywhere in Netra.
Evaluators
Added five image evaluators:- Image Analysis
- Image Editing Quality
- Image Helpfulness
- Image-Text Coherence
- Image-Text Reference
Python SDK v0.1.77
- Custom metrics: Added an SDK utility for recording custom metrics. See Custom metrics.
PlatformObservabilityMCPNew featureImprovement
Prompt Management, Netra MCP, and on-premise deployment
Prompt Management
- Prompt Management: Released Prompt Management in all regions. See Prompt Studio.
Developer tools
- Netra MCP: Launched Netra MCP, so you can explore traces and run root-cause analysis from your coding agent.
- Netra Skills: Launched Netra Skills for guided trace, evaluation, and simulation setup in your IDE.
Observability
- Cleaner trace timeline: Reduced irrelevant spans in the trace timeline.
- Rich chat previews: Chat previews now render Markdown and JSON.
Deployment
- On-premise deployment: Netra is now available for on-premise deployment.
Python SDK v0.1.76
- Bug fixes: Minor bug fixes.
SimulationsEvaluationsDatasetsDashboardsPlatformPython SDKTypeScript SDKNew featureBug fix
Simulations, multi-turn evaluation, and export management
Simulations and evaluations
- Simulations and multi-turn evaluation: Released simulations and multi-turn evaluation in the US and EU regions.
- Datasets from traces: Fixed issues when creating datasets and evaluations from traces.
Data exports
- Delete exports: You can now delete data exports from the Netra UI.
Dashboards
- Audio and character metrics: Custom widgets now support Audio Duration and Character Count metrics. See Custom dashboards.
SDKs
- Python SDK v0.1.67 and TypeScript SDK: Added simulation support. See the Python and TypeScript simulation references.
PlatformEvaluationsDashboardsObservabilityPython SDKNew featureImprovementBug fix
Netra v3.1.0, evaluations, data exports, and dashboard APIs
Netra v3.1.0
- Project switcher: Added a project switcher so you can move between projects quickly.
- Reorganized navigation: Reorganized settings, profile, and organization navigation.
- Organization owner permissions: Added permissions for organization owners. See Organization settings.
- Sign-in fixes: Fixed login and session issues.
Subscriptions and billing
- Plan limits: Plans now include limits on user invites and data retention.
- Overage estimates: Added an estimate of overage charges.
- Billing portal: Added a Stripe customer portal to manage your subscription.
- Subscription lifecycle: Added handling for cancellations, renewals, and billing cycles.
Evaluations
- Evaluations: Released evaluations in the US region.
- Automatic evaluation: Production traces can now be evaluated automatically, with scores and reasoning for Coherence, Factual Accuracy, and Toxicity.
Data exports
- Scheduled exports: Released scheduled data exports. See Project settings.
Dashboards and APIs
- Public dashboard APIs: Released public dashboard APIs in the US and EU regions.
- Session APIs: Released session APIs. See Sessions.
Python SDK v0.1.61
- Dashboard query utility: Added an SDK utility for the dashboard API. See the dashboard query reference.
Python SDK v0.1.63
- Bug fixes: Minor bug fixes.