# TripleCloud

> LLM and AI agent observability for AI pipelines — quality, cost and stability control for teams shipping LLMs in production. The LLM observability platform powered by the open-source IceGate engine.

TripleCloud gives engineering teams online LLM evaluation, LLM monitoring, token-cost accounting and budgets, and distributed tracing, SLOs and alerting for production LLM systems. The open-source IceGate engine stores all telemetry — write-ahead log, Parquet files and catalog metadata — in S3-compatible object storage; TripleCloud is the LLM observability platform built on it.

## Product

- [TripleCloud (marketing site)](https://triplecloud.tech): What it is, features and pricing.
- [Control plane app](https://triplecloud.app): Dashboards, conversations, SLOs, alerts and cost.

## Engine

- [IceGate](https://icegate.tech): The open-source observability data lake engine that powers TripleCloud, built by the same team. OpenTelemetry ingest into Apache Iceberg on object storage, with Loki/Prometheus/Tempo-compatible read APIs plus Arrow Flight SQL, and no dedicated database. Apache 2.0 — self-hostable.
- [IceGate documentation](https://docs.icegate.tech): Installation, configuration, query APIs and architecture reference.
- [IceGate source (GitHub)](https://github.com/icegatetech/icegate): Rust implementation, issues and contributing guidelines.

## Writing

- [Blog](https://blog.triplecloud.tech): Articles on LLM observability, evals and running LLMs in production.
- [OpenTelemetry GenAI semantic conventions: how to instrument an LLM agent](https://blog.triplecloud.tech/posts/instrument-llm-agent-opentelemetry): OpenTelemetry GenAI semantic conventions, hands-on: auto-instrumentation, agent and tool spans, token usage and cost, and where to send the traces.
- [LLM monitoring and observability: the AI tax on ordinary telemetry](https://blog.triplecloud.tech/posts/the-ai-tax-on-telemetry): LLM monitoring and observability tools bill every span at the premium AI rate, even ordinary HTTP and database spans. The fix: price AI data as AI data.
- [Can you build observability ingestion on S3 alone - no Kafka, no disks, no coordination layer?](https://blog.triplecloud.tech/posts/observability-ingestion-on-s3): An open source observability stack running on S3 alone - no Kafka, no local disks, no coordination service. Data, WAL and Iceberg catalog all live in S3.

## Capabilities

- LLM-as-judge and human-feedback quality evaluation
- Hallucination and PII-leak detection
- Token cost rollups, budgets and projected spend
- Distributed traces across agents and tool calls
- SLOs with burn-rate and error budgets
- Alerting with on-call paging