# Netra > Netra is the AI agent observability, evaluation, and simulation platform. Trace every LLM call and agent decision, evaluate quality before and after every release, and simulate real-world behavior — in one place. Netra is built for non-deterministic AI agent workflows, not retrofitted from traditional APM tools. It is built on OpenTelemetry standards and organized around four pillars: Observability, Evaluation, Simulation, and Monitoring & Alerting. Built by KeyValue Software Systems, which deployed 25+ AI agents in production before building Netra. --- ## Marketing Site (getnetra.ai) Authoritative product and company pages on the marketing website: - [Homepage](https://getnetra.ai/): AI observability, evaluation, and simulation platform overview - [Observability](https://getnetra.ai/platform/observability/): Trace every LLM call, tool execution, and agent decision - [Evaluation](https://getnetra.ai/platform/evaluation/): Score agent quality with built-in and custom evaluators - [Simulation](https://getnetra.ai/platform/simulation/): Multi-turn agent simulation with configurable personas - [Agent Insights](https://getnetra.ai/platform/agent-insights/): Know what is breaking and why in production agents - [Pricing](https://getnetra.ai/pricing/): Free, Pro ($39/mo), and Enterprise plans - [About Netra](https://getnetra.ai/about-us/): Company story and principles - [Contact / Book a Demo](https://getnetra.ai/contact-us/): Sales and demo requests - [Privacy Policy](https://getnetra.ai/privacy-policy/) - [Terms of Service](https://getnetra.ai/terms-of-service/) - [Blog](https://getnetra.ai/blog/): Product updates and engineering articles AI discovery files: - [llms.txt](https://getnetra.ai/llms.txt) — this file - [ai.txt](https://getnetra.ai/ai.txt) — AI crawler manifest - [Sitemap](https://getnetra.ai/sitemap.xml) --- ## Getting Started - [What is Netra?](https://docs.getnetra.ai/netra): Platform overview, architecture, and core concepts - [Overview](https://docs.getnetra.ai/quick-start/Overview): Quickstart index — start here - [Quickstart: Tracing](https://docs.getnetra.ai/quick-start/QuickStart_Tracing): Instrument your app in minutes with 2–3 lines of code - [Quickstart: Evaluation](https://docs.getnetra.ai/quick-start/QuickStart_Evals): Run your first evaluation against production data - [Quickstart: Simulation](https://docs.getnetra.ai/quick-start/QuickStart_Simulation): Run your first multi-turn agent simulation - [Quickstart: Alerts](https://docs.getnetra.ai/quick-start/QuickStart_Alerts): Set up your first alert rule - [FAQs](https://docs.getnetra.ai/FAQs/FAQs): Common questions about setup, features, and pricing --- ## Observability & Tracing - [Traces Overview](https://docs.getnetra.ai/Observability/Traces/overview): End-to-end request tracing — every LLM call, tool invocation, and decision path, with latency (P50/P95/P99), token usage, and cost per request - [Tenants](https://docs.getnetra.ai/Observability/Tenants): Multi-tenant trace isolation; tag traces by `netra.tenant_id` for per-customer visibility, cost tracking, and alerting - [Agents](https://docs.getnetra.ai/Observability/Agents): Per-agent observability views - [Users](https://docs.getnetra.ai/Observability/Users): Per-user trace grouping and analytics - [Sessions](https://docs.getnetra.ai/Observability/Session): Session-level trace aggregation across multi-turn interactions Tracing methods: auto instrumentation (zero code changes), decorators (`@agent`, `@task`, `@workflow`), and manual tracing for full control. --- ## Evaluation - [Evaluation Overview](https://docs.getnetra.ai/Evaluation/Evaluation-overview): Test every change against production data before it reaches users; catch regressions before they become incidents - [Evaluators](https://docs.getnetra.ai/Evaluation/Evaluators): LLM-as-Judge evaluators for subjective quality; code evaluators for deterministic checks (JSON schema, regex, custom logic); pre-built library covers coherence, factual accuracy, toxicity, helpfulness, tool call correctness, policy compliance - [Datasets](https://docs.getnetra.ai/Evaluation/Datasets): Create test datasets from real production traces with one click, or build manually - [Test Runs](https://docs.getnetra.ai/Evaluation/TestRuns): Point-in-time snapshots comparing expected vs. actual outputs; track quality trends across releases Auto-evaluation: once live, every trace is automatically scored on coherence, factual accuracy, and toxicity — no manual triggers, no sampling. --- ## Simulation - [Simulation Overview](https://docs.getnetra.ai/Simulation/Simulation-overview): Run full multi-turn conversational simulations with configurable personas and goals before real users ever see the agent - [Simulation Evaluators](https://docs.getnetra.ai/Simulation/Evaluators): 8 pre-configured evaluators covering guideline adherence, conversation completeness, goal fulfillment, factual accuracy, conversation memory, and information elicitation - [Simulation Datasets](https://docs.getnetra.ai/Simulation/Datasets): Scenario and persona datasets for simulation runs - [Simulation Test Runs](https://docs.getnetra.ai/Simulation/TestRuns): Run, review, and compare simulation results Each simulation defines a scenario goal, a user persona (frustrated / confused / friendly / neutral / custom), optional fact-checking criteria, and a max turn limit to detect infinite loops. --- ## Monitoring & Alerting - [Alert Rules](https://docs.getnetra.ai/Alert-rules/Alert-rules): Proactive real-time alerts on cost overruns, latency degradation, error rate spikes, and SLA violations; notification via email or Slack - [Custom Dashboard](https://docs.getnetra.ai/Dashboard/Custom-dashboard): Build tailored analytics views with 6 chart types and 4 pre-built templates (Cost Monitoring, Performance Monitoring, Quality & Compliance, Tenant Analytics); metrics include Error Rate, Latency P50–P99, Request Count, Total Cost, Total Tokens, Time to First Token --- ## Prompt Studio - [Prompt Studio](https://docs.getnetra.ai/prompt-studio): Centralized prompt management — build, version, test, and deploy prompts without code redeployment; immutable published versions with GitHub-style diffs; labels (`production`, `staging`, `latest`) for controlled rollouts --- ## MCP Server - [MCP Overview](https://docs.getnetra.ai/mcp/overview): Query Netra observability context directly from AI coding assistants (Cursor, Claude Code, Windsurf, Antigravity, Codex) MCP endpoints: US `https://api.getnetra.ai/mcp` · EU `https://api.eu.getnetra.ai/mcp` Available tools: `netra_get_trace_by_id`, `netra_query_traces` --- ## SDKs & Integrations - [SDK Overview](https://docs.getnetra.ai/sdk-reference/sdk/overview): Python and TypeScript SDKs - [Python SDK Reference](https://docs.getnetra.ai/sdk-reference/sdk/python): `pip install netra-sdk` — [PyPI](https://pypi.org/project/netra-sdk/) · [GitHub](https://github.com/KeyValueSoftwareSystems/netra-sdk-py) - [TypeScript SDK Reference](https://docs.getnetra.ai/sdk-reference/sdk/typescript): `npm install netra-sdk` — [npm](https://www.npmjs.com/package/netra-sdk) · [GitHub](https://github.com/KeyValueSoftwareSystems/netra-sdk-js) - [Integrations Overview](https://docs.getnetra.ai/Integrations/overview): Full list of supported providers, frameworks, and databases **LLM providers (14+):** OpenAI, Anthropic Claude, Google Gemini, Vertex AI, AWS Bedrock, Mistral, Groq, Cohere, Ollama, Together AI, Replicate, Hugging Face, Aleph Alpha, IBM watsonx **AI frameworks (12+):** LangChain, LangGraph, LlamaIndex, CrewAI, Pydantic AI, LiteLLM, Haystack, DSPy, Google ADK, Cerebras, AutoGen, MCP **Vector databases (8):** Pinecone, ChromaDB, Qdrant, Weaviate, Milvus, LanceDB, Redis, Marqo **Export targets:** Prometheus, Datadog, New Relic --- ## Pricing - [Pricing Page](https://getnetra.ai/pricing/): Free ($0, 5k spans/month), Pro ($39/month, 200k spans), Enterprise (custom). No credit card required for Free. Enterprise adds isolated database, SSO, on-premise deployment, and dedicated support. --- ## Security & Compliance SOC 2 Type II · HIPAA · GDPR. PII detection and masking. Prompt injection detection. Encryption in transit and at rest. RBAC. Configurable data retention. Data regions — US: `https://api.getnetra.ai/telemetry` (us-east-1) · EU: `https://api.eu.getnetra.ai/telemetry` (eu-central-1). Regions are strictly separated; no cross-region data sharing. --- ## Migrations - [Helicone to Netra](https://docs.getnetra.ai/Migrations/Helicone): Migration guide for teams moving from Helicone --- ## Cookbooks & Guides - [Cookbooks](https://docs.getnetra.ai/Cookbooks/observability/tracing-rag-pipeline): End-to-end examples including RAG pipeline tracing and more --- ## Customers > "Netra turned our ad-agent stack from a black box into a clear control center. Issues that once took hours to debug now take minutes." — Sumukh Avadhani, CTO, Pencil (1M+ spans/day) > "Being able to evaluate and trust what our agents produce before anything goes out has changed the speed of our shipping." — Sathya, CTO, Aartha AI > "Netra helped us catch broken loops early, cut incident investigation time by over 50%, and reduce AI spend per customer by about 30%." — Prasen, CEO, Axari Current customers: Pencil, Aartha AI, Axari --- ## Key Links - Website: https://getnetra.ai/ - Documentation: https://docs.getnetra.ai/ - Pricing: https://getnetra.ai/pricing/ - Blog: https://getnetra.ai/blog/ - About: https://getnetra.ai/about-us/ - Sign up: https://app.eu.getnetra.ai/signup - Book a demo: https://getnetra.ai/contact-us/ - US app: https://app.getnetra.ai - EU app: https://app.eu.getnetra.ai - Sales: sales@getnetra.ai - LinkedIn: https://www.linkedin.com/company/getnetraai/ - Twitter / X: https://x.com/getnetraAI