Built for AI Infrastructure

Intelligence Behind Every Request

Route AI requests, observe agent activity, and understand every interaction through one unified gateway.

Explore Kyurem

A2A Gateway & Observability Platform

One Gateway. Complete Visibility.

A unified layer to route, monitor, and understand AI agent interactions across your infrastructure.

Protocol at the wire’s edge

Built in Go to transparently proxy A2A v1.0 traffic over JSON-RPC, HTTP+JSON, and SSE streaming so tracing, trust, and resilience stay out of individual agent code.

Non-blocking telemetry

Stateless request streaming paired with a bounded OpenTelemetry export queue ensures spans drop under extreme load before upstream traffic is ever delayed.

Isolated plane boundaries

Dedicated listeners separate the A2A data path (:8080), operations endpoints (:8081), and bearer-authenticated management API (:8082).

Core Capabilities

Built for multi-agent traffic at the edge.

AI Gateway

Centralized routing by agent name or alias across JSON-RPC, HTTP+JSON, and SSE streams. Handles agent card rewriting, caching, W3C trace context propagation, and graceful draining.

Observability

Monitor requests, OpenTelemetry handoff spans, task traces, latency, and upstream errors through a bounded OTLP queue backed by in-memory ring or ClickHouse storage.

Agent Graph

Visualize caller-to-upstream relationships and hop topologies across distributed tasks so multi-agent handoffs remain traceable end to end.

Cost Intelligence

Track input and output token usage and calculate request costs across tasks and spans. When upstream providers omit usage, cost remains explicitly unknown rather than masked as zero.

Policy & Audit

Authenticate callers via bearer tokens or mTLS, verify agent card signatures against operator-configured keys, enforce deny-overrides rules, and inspect bounded audit records.

Circuit Breakers

Protect upstream agents with bounded retry budgets, same-binding failover on 502/503/504 idempotent calls, and per-upstream circuit breakers that trip after consecutive failures.

Management Dashboard

Every task, span, and failover event in one place.

Kyurem’s React dashboard connects to the dedicated management API to turn raw OTLP handoff traces into diagnosable task timelines.

  • Task monitoring Duration, status, agent count, and critical-path root cause
  • Trace and span inspection Span trees, waterfall timelines, and gen_ai.* / kyurem.* attributes
  • Agent relationships Caller-to-callee handoffs across Go and Python A2A agents
  • Latency and error visibility kyurem.retry and kyurem.failover span events surfaced per hop
  • Cost analytics Per-task token and cost rollups with explicit unknown-usage accounting
Kyurem management dashboard displaying a task detail view with span tree, timeline waterfall, cost breakdown, and retry/failover span events.

Architecture

How Kyurem routes and observes.

Agents point at the gateway without SDK rewrites while telemetry and control planes run out of band.

Ingress

AI Applications

Calling agents via JSON-RPC, HTTP+JSON, or SSE

Data Path :8080

Kyurem Gateway

Caller auth · Policy · Router · Retry & Circuit Breaker

Upstreams

Agents / AI Providers

Go, Python, or external A2A-compliant services

1. Span Builder

Router and resilience layers emit one OpenTelemetry handoff span into a bounded async queue.

2. OTLP Collector & Store

Traces export via OTLP to an OpenTelemetry Collector backed by ClickHouse or an in-memory ring.

3. Ops & Management

Prometheus scrapes :8081/metrics while the dashboard queries the :8082 management API.

Developer Experience

Built for developers. Designed for visibility.

Configure Kyurem with a single strict YAML file validated at startup—unknown keys are rejected early and scalar fields can be overridden with KYUREM_* environment variables.

  • Validate deployments before launch with kyurem validate
  • Typed OpenAPI contract at api/openapi.yaml for the management API
  • Local Go and Python demo walkthroughs for tracing and failover
View on GitHub Configuration Docs
Quick start & validation Go 1.26+ · YAML 
# Validate configuration strictly before serving
go run ./gateway/cmd/kyurem validate --config config/kyurem.example.yaml

# Build and run the Go demo walkthrough
make build build-demo
make run-demo

# Failover walkthrough with Python agents + dashboard
make python-install dashboard-install
make demo-m6
Data path :8080 Ops :8081 Management API :8082

See everything behind every request.

Bring visibility and control to your AI infrastructure with Kyurem.

Explore Kyurem on GitHub