Unified observability and security for AI agents.
Website · Docs · Cloud App · Core Repo · Blog
Agents are moving from weekend demos into production — calling tools, spending money, touching real systems, running unattended. Traditional APM tools were built for predictable request/response services, not non-deterministic, multi-step, multi-agent workflows.
Splyntra exists so teams can see exactly what their agents did, whether it was safe, and what it cost — all on a single, OpenTelemetry-native pipeline. No stitching together a tracing tool, a security tool, and a spreadsheet for cost tracking. One data model, viewed from every angle that matters.
We build it source-available first — self-host the full core for free, forever — with a managed cloud on top for teams who'd rather not run the infrastructure themselves.
| Pillar | What it does |
|---|---|
| 🔭 Observability | Distributed traces of agent runs, tool calls, and LLM decisions, plus agent replay, trace-correlated logs, and cost/latency analytics per run, model, or project |
| 🛡️ Security | One risk score per run — combining secret/PII detection, prompt-injection detection, content moderation, and tool-guard — mapped to the exact span it happened on |
| 🧪 Evaluation | Dataset management, scorers (exact/rule/tool-call/latency/cost/groundedness), version-over-version regression, and a CI gate via the splyntra CLI |
| 🔐 Governance | Hash-chained audit ledger, self-service delegation with spend limits, and a policy engine (RBAC/ABAC/ReBAC) |
| 🪪 Agent identity | Registered agents with scoped credentials, cross-agent trust policies, and OIDC federation |
| Repo | What it is |
|---|---|
splyntra |
The core: OTel collector, dashboard, detectors, evaluation engine, and infra — self-hostable via docker compose up |
sdks/python |
pip install splyntra — decorators, auto-instrumentation, redaction |
sdks/typescript |
npm install @splyntra/sdk — function wrappers, decorators, CLI |
git clone https://github.com/splyntra/splyntra
cd splyntra
docker compose up -dpip install splyntra
# or
npm install @splyntra/sdkThen open http://localhost:3000/traces and watch your first agent run come in
with a risk score attached. Full walkthrough: Quickstart.
We welcome issues, discussions, and pull requests on the core repo:
- Start with
CONTRIBUTING.mdfor local dev setup, Conventional Commits, and how releases work. - Contributions require signing our CLA.
- Found a security issue? Please don't open a public issue — see
SECURITY.mdfor responsible disclosure.
The server core, detectors, and dashboard are Functional Source License 1.1 — free for any use inside your company at any scale, just not resellable as a competing product; each release becomes Apache-2.0 two years later. The Python and TypeScript SDKs are plain Apache-2.0. Governance, identity/SSO, and the control plane are commercial, in a separate private repo. Full breakdown in LICENSING.md.
- Questions / feedback: open a Discussion on the core repo
- Docs: splyntra.com/docs
- Updates: splyntra.com/blog
Built on OpenTelemetry. Boring infrastructure (Go + ClickHouse + Postgres), novel product.