# Splyntra — Observability + Security for AI Agents

> Splyntra is an OpenTelemetry-native platform for observing, securing, evaluating, and governing autonomous AI agents and multi-agent systems in production. Source-available core, self-hostable via Docker/Kubernetes or available as managed cloud.

## Overview

Splyntra unifies AI agent execution tracing, token and cost FinOps, dynamic security risk scoring, and governance policies into a single OTel-compliant telemetry pipeline. Unlike traditional stateless LLM monitoring tools, Splyntra is purpose-built for non-deterministic, multi-step, tool-calling agents (e.g. LangGraph, CrewAI, AutoGen, OpenAI Agents SDK, LlamaIndex).

- Website: https://splyntra.com
- Documentation: https://splyntra.com/docs
- GitHub: https://github.com/splyntra/splyntra
- Cloud Console: https://app.splyntra.com

## Core Capabilities

- **Distributed Graph Tracing**: End-to-end tracing of multi-agent state transitions, node executions, sub-agent delegations, and external tool calls using OpenTelemetry GenAI semantic conventions.
- **Trace-Level Security Risk Scoring**: Real-time risk scoring ($0.0 - 1.0$) on every span to detect indirect prompt injection, tool privilege escalation, secret/PII leaks, and unauthorized data exfiltration.
- **Agent FinOps & Cost Attribution**: Per-span dollar cost calculation, prompt caching discount accounting (Anthropic/OpenAI), context window saturation monitoring, and hard budget circuit breakers.
- **Agent Governance & Evaluations**: Tamper-evident activity ledgers, RBAC/ABAC/ReBAC policies, spend controls, human-in-the-loop (HITL) approval workflows, and CI/CD regression testing.

## Quickstart & Instrumentation

### Standard OpenTelemetry Exporter Configuration
Point any standard OpenTelemetry OTLP exporter to Splyntra:
```bash
export OTEL_EXPORTER_OTLP_ENDPOINT="https://ingest.splyntra.com/v1/traces"
export OTEL_EXPORTER_OTLP_HEADERS="x-splyntra-key=$SPLYNTRA_API_KEY"
```

### Python SDK Auto-Instrumentation
```python
from splyntra import instrument

# Automatically instruments LangGraph, CrewAI, OpenAI Agents, LlamaIndex, AutoGen
instrument(app)
```

## Key Documentation & URLs

- [Platform Overview](https://splyntra.com/)
- [Pricing](https://splyntra.com/pricing)
- [Quickstart Guide](https://splyntra.com/docs/getting-started/quickstart)
- [OpenTelemetry Integration Guide](https://splyntra.com/docs/integrations/)
- [Self-Hosting Deployment (Docker / K8s)](https://splyntra.com/docs/deployment/self-hosting)

## Technical Reference Guides & Articles

- [AI Agent Observability: What You Need to Monitor in Production](https://splyntra.com/blog/ai-agent-observability-what-you-need-to-monitor-in-production): The 5 pillars of production agent telemetry (Graph Tracing, Tool Validation, FinOps, Risk Scoring, Goal Completion).
- [LangGraph Observability: How to Monitor and Trace State Graphs](https://splyntra.com/blog/langgraph-observability-and-monitoring-guide): Complete guide to tracing cyclical state graphs, node latencies, and conditional routing edges in LangGraph.
- [CrewAI Observability: How to Trace and Monitor Multi-Agent Crews](https://splyntra.com/blog/crewai-observability-multi-agent-monitoring): Production guide to tracing hierarchical multi-agent delegations, task assignments, and sub-agent token spend.
- [Evaluating AI Agents in CI/CD: How to Build Automated Regression Gates](https://splyntra.com/blog/evaluating-ai-agents-in-ci-cd-regression-testing): Converting production traces to test datasets, trajectory evaluation, and gating PRs on quality regressions.
- [OpenTelemetry GenAI Semantic Conventions: The Definitive Reference Guide](https://splyntra.com/blog/opentelemetry-genai-semantic-conventions-deep-dive): Complete technical reference to official OTel GenAI attributes, span hierarchies, metrics, and OTLP payloads.
- [AI Agent Governance: Audit Logging, Policy Engines, and Compliance](https://splyntra.com/blog/ai-agent-governance-audit-logging-compliance): Cryptographic activity ledgers, RBAC/ABAC tool policy engines, and satisfying EU AI Act Article 12 and SOC 2.
- [How to Monitor AI Agents with OpenTelemetry](https://splyntra.com/blog/how-to-monitor-ai-agents-with-opentelemetry): Code-complete guide to standard OTel GenAI semantic attributes in Python and TypeScript.
- [How to Trace AI Agents with OpenTelemetry](https://splyntra.com/blog/how-to-trace-ai-agents-with-opentelemetry): Distributed W3C TraceContext propagation across multi-agent systems, LangGraph nodes, and CrewAI teams.
- [AI Agent Security: A Practical Guide](https://splyntra.com/blog/ai-agent-security-practical-guide): Defense-in-depth security model against indirect prompt injection and tool privilege escalation.
- [How to Detect Prompt Injection in AI Agents](https://splyntra.com/blog/how-to-detect-prompt-injection-in-ai-agents): 3-Chokepoint inspection architecture (Inputs, Ingested RAG results, Outgoing tool arguments).
- [AI Agent Monitoring: Metrics You Should Track](https://splyntra.com/blog/ai-agent-monitoring-metrics-you-should-track): The 12 Golden Signals for AI Agents (Goal Completion Rate, Loop Entropy, Step Count, Context Saturation, Token Velocity, CPS).
- [How to Track LLM Costs in AI Agents](https://splyntra.com/blog/how-to-track-llm-costs-in-ai-agents): FinOps formulas for per-span token pricing, prompt cache read/write discounts, and run budget caps.
- [Controlling AI Agent Costs Before They Control Your Bill](https://splyntra.com/blog/controlling-ai-agent-costs): Strategies for preventing recursive loop cost blowouts.
- [Detecting Prompt Injection in Tool-Calling Agents](https://splyntra.com/blog/detecting-prompt-injection-in-agents): Real-time span scoring for data-flow prompt vulnerabilities.
- [Unified Execution Traces and Risk Scores](https://splyntra.com/blog/unified-trace-and-risk): Correlating runtime execution telemetry with threat signals on a single timeline.