Traces that fuel
Confluent Intelligence.
Confluent tells you what your agent knew. tapes tells you what your agent did. Zero-instrumentation observability for AI agents in production, streamed to Confluent Cloud.
The Problem
AI agents are shipping to production across every industry. They make LLM calls, invoke tools, access data, and make autonomous decisions. But when something breaks or compliance asks what happened, there is no record.
Paper Compute
Open source infrastructure for AI agents in production.
Zero-instrumentation observability layer that captures every LLM call, tool invocation, and execution path at the network layer. No SDKs. No code changes. One command.
Team dashboard for governing agents at scale. Org-scoped session management, cost tracking, anomaly detection, and session comparison.
Where Confluent Fits
Paper Compute uses Confluent as the streaming backbone for agent telemetry in production.
The event backbone inside Paper Cloud. Every agent action captured by tapes is published to Kafka topics as structured traces—a single durable stream of what every agent did, when, and why.
Real-time anomaly detection purpose-built for agent failure modes. Detects stuck loops, token spikes, behavioral drift, and prompt injection attempts.
Paper Cloud publishes structured agent telemetry to Confluent Cloud via a Kafka producer. Traces stream directly to your existing Confluent environment. No pipeline stitching.
Confluent's Real-Time Context Engine feeds live context back into agents. tapes captures what agents do with that context. This context feedback loop is a Paper Compute integration pattern built on top of Confluent primitives. Context in, telemetry out—a closed observability loop.
Real-Time Anomaly Detection
Purpose-built for agent failure modes. Every anomaly follows a Detect → Understand → Act pattern so teams can go from alert to resolution in one flow.
Integration Roadmap
Turn raw agent telemetry into queryable tables so compliance and security teams can audit agent behavior without engineering involvement.
Generate synthetic agent traffic to test detection pipelines before deploying to production.
Feed tapes anomaly data back into the context engine so agents can self-correct based on historical failures.
The Opportunity
Confluent customers deploying agents in production need answers to two questions:
Together, Confluent and Paper Compute give enterprises full-stack observability for agent workloads. Regulated industries—financial services, healthcare, defense—that already require audit trails for data pipelines will require the same for the agents operating on that data.
Confluent was acquired by IBM in 2025. For enterprise buyers, this means Confluent Cloud is now backed by IBM’s global support, compliance certifications, and procurement infrastructure. Paper Compute’s integration targets Confluent Cloud APIs and is unaffected by the change in ownership.