01
Models
Every model call across every service — provider, model name, version and routing — mapped automatically.
Carraggon Cover · AI Observe overview
Full visibility into how your services use AI.
Models, cost, latency, prompts, agent behavior and tool execution — observed in one place. All data stays in your infrastructure with BYOC (Bring Your Own Cloud); Carraggon never processes your AI data outside your environment.
✓ eBPF auto-detection, zero code changes✓ OpenTelemetry agent-level tracing★ Your cloud, your data
What you see
Every AI call, every agent run, accounted for.
02
Cost
Token spend attributed per call, service, team and feature, so runaway cost is visible the hour it starts.
03
Latency
Time-to-first-token, streaming duration and total call latency, broken down by model and endpoint.
04
Prompts
Full prompt and completion payloads, retained under your own redaction and access policy.
05
Agent behavior
Multi-step agent runs traced end to end — reasoning steps, retries, loops and dead ends.
06
Tool execution
Every tool and function call an agent makes, with arguments, results, errors and duration.
Two telemetry sources
Instant per-call visibility, plus the full agent-level picture.
Zero code changes
eBPF auto-detection
Carraggon Cover detects AI traffic at the kernel level, giving immediate per-call visibility the moment it is deployed. No SDKs, no redeploys, no instrumentation work.
- Per-call visibility from minute one
- Captures calls from any language or framework
- No application changes or redeploys
- Sees traffic your SDKs miss
Full agent picture
OpenTelemetry instrumentation
Standard OTel spans give the agent-level story: how a run was decomposed, which tools fired, what came back and where a run went wrong.
- End-to-end agent run traces
- Tool and function call spans
- Vendor-neutral, standards-based
- Correlated with your existing traces
Bring Your Own Cloud
Your prompts never leave your environment.
Data never leaves
Telemetry is stored and processed inside your own cloud account. Carraggon never processes your AI data outside your environment.
You own retention
Set your own retention windows, redaction rules and access controls — prompts are sensitive data and stay under your policy.
Compliance-friendly
Residency, tenancy and audit requirements are satisfied by construction, not by a vendor promise.
Outcomes
0
lines of code changed to get first AI visibility
100%
of AI calls captured, not a sampled subset
BYOC
all telemetry stays inside your own cloud
Minutes
from install to per-call model and cost data