Observability + security
for AI agents
Splyntra traces every agent run and attaches a risk score — so you can see what your agents did, what it cost, and whether anything risky happened. One OpenTelemetry-based pipeline instead of five disconnected tools.
Free forever · No credit card required · FSL-1.1 source-available core
Works with your agent stack
Stop stitching five tools together
Tracing in one tool, cost in a spreadsheet, security in another, evaluations in a notebook, governance nowhere. Splyntra puts execution, cost, quality, and risk on a single pipeline — built on open standards, shipped as a source-available core with a managed cloud on top.
Every agent run, traced end to end
OpenTelemetry-native tracing for agents, LLM calls, tool calls, and retrieval — with cost attached to every step. See what ran, how long it took, and what it cost, in one timeline.
- ✓Distributed traces across agents, tools, and models
- ✓Cost analytics per run, per model, and per project
- ✓Metrics, logs, and retrieval quality on the same data
A risk score on every trace
Detection runs inside the pipeline, not as a bolt-on. Secrets and PII are redacted by default before anything is stored, prompt-injection attempts are flagged, and each run carries a single risk score.
- ✓Secret + PII detection, redacted by default
- ✓Prompt-injection detection in inputs and tool calls (beta)
- ✓Unified risk view — execution and security side by side
Ship changes without regressions
Turn traces into datasets, score them, and fail CI when quality drops. Then govern what agents are allowed to do — with an activity ledger, spend controls, approvals, and a policy engine on the same pipeline.
- ✓Datasets, scorers, and CI regression gates
- ✓Tamper-evident activity ledger + spend controls & approvals
- ✓Policy engine (RBAC/ABAC/ReBAC), agent identity, SSO/SAML/SCIM
Live in five minutes
Instrument your agent with the SDK, or point any OpenTelemetry exporter at Splyntra. Self-host the source-available core with a single docker compose up, or start free on the managed cloud.
# Any OpenTelemetry exporter works — point it at Splyntraexport OTEL_EXPORTER_OTLP_ENDPOINT="https://ingest.splyntra.com"export OTEL_EXPORTER_OTLP_HEADERS="x-splyntra-key=$SPLYNTRA_API_KEY" # …or use the SDK in your agentfrom splyntra import instrumentinstrument(app) # LangGraph, CrewAI, OpenAI Agents, … # every run now shows up as a trace — with cost and a risk scoreAuditor-ready by design
Generate evidence reports mapped to the frameworks your auditors ask about — straight from the activity ledger and policy data, no manual screenshots.
Source-available, your data
The foundation is source-available (FSL-1.1) and self-hostable — run it with a single docker compose up and keep every byte on your own infrastructure. Enterprise governance and the fully managed cloud are built on the same core — add them when you need them, without re-platforming.
Self-host, single org — free forever.
- Tracing, metrics & logs
- Cost analytics
- Secret + PII detection · prompt-injection
- Evaluation & CI gates
Self-host add-ons or fully managed.
- Governance: ledger, policies, spend controls
- Agent identity & trust
- Compliance reports (SOC 2 / EU AI Act / NIST)
- SSO / SAML / SCIM · multi-tenant orgs
Frequently asked questions
Everything you need to know about AI agent observability, OpenTelemetry tracing, and threat defense.
What is AI agent observability and how is it different from LLM monitoring?
AI agent observability monitors non-deterministic, multi-step, tool-calling agents across graph state transitions, external API parameters, and sub-agent delegations. Unlike single-turn LLM chatbots, autonomous agents require distributed OpenTelemetry tracing, FinOps loop circuit breakers, and real-time security risk scoring on every span.
How does Splyntra trace and monitor AI agents in production?
Splyntra uses standard OpenTelemetry GenAI semantic conventions to record every reasoning step, model completion, and tool invocation on one continuous timeline. You can auto-instrument frameworks using the Splyntra SDK or point any standard OpenTelemetry exporter directly to Splyntra.
Which AI agent frameworks does Splyntra support?
Splyntra natively integrates with LangGraph, CrewAI, OpenAI Agents SDK, LlamaIndex, AutoGen, Dify, and any custom agent pipeline exporting OTLP traces.
How does Splyntra detect prompt injection in autonomous agents?
Splyntra inspects data flows at three runtime chokepoints: model inputs, ingested tool results (e.g. web pages, PDFs), and outgoing tool call arguments. Each span receives a real-time risk score (0.0 to 1.0) to detect behavioral anomalies and intercept unauthorized actions before execution.
How does Splyntra prevent runaway LLM costs from infinite loops?
Splyntra attributes token spend down to the exact span, accounts for Anthropic/OpenAI prompt cache discounts, and allows you to enforce hard spend and step-count circuit breakers to immediately abort runaway loops.
Can Splyntra be self-hosted on private infrastructure?
Yes. Splyntra's core is source-available and easily self-hosted with Docker Compose or Kubernetes, ensuring your sensitive customer data and traces never leave your perimeter.
See your agents clearly
Trace, evaluate, secure, and govern your AI agents on one pipeline. Start free — no credit card required.