Hallucination Detection
in/guard/out vs Galileo Luna
Galileo’s Luna family – small fine-tuned evaluation models – makes continuous quality scoring affordable: low latency, a fraction of judge-model cost, wired into Galileo’s evaluation and observability platform. Cisco bought Galileo in April 2026 for an undisclosed sum, and it now ships as Splunk Agent Observability: evaluation, tracing, usage and cost tracking, and real-time guardrails on agent actions, alongside the rest of the Splunk stack.
Those guardrails are built on scores – a model’s judgement of how good an output looks, checked against a threshold. in/guard/out starts from facts about the run: whether a figure matches its source, whether a payee appears in what the agent read, whether the next call fits the budget it reserved. Those checks cost nothing, need no threshold, and – where a safe fix exists – repair the output instead of only stopping it.
Side by side
| Capability | Galileo Luna | in/guard/out |
|---|---|---|
| What it is | Small eval models inside an observability platform – now part of Splunk | A runtime control plane – checks with enforcement authority |
| First line of defense | A small-model inference per metric | Zero-token deterministic checks before any model spends |
| What a guardrail acts on | A model’s score against a threshold | The run’s own facts – sources read, budget reserved, values screened – plus opt-in model tiers |
| On a failed check | Block or alert | Repair, block, or flag – the drifted figure is corrected, the JSON fixed, before the response ships |
| Scope | Hallucination/quality metrics, agent tracing | Output checks plus PII round-trip, format repair, injection, tool policy, budgets |
| Agent runs | Traced, scored, guarded per action | Governed across the run: hard budgets, action grounding, taint, sequence policies |
| Integration | SDK instrumentation into the platform | One base-URL change; no app code |
| Cost accounting | Usage and cost tracking | Every guard-model call priced into the request and the run – the guardrail tax is a number |
Based on Galileo Luna’s publicly available information. We’ve done our best to keep it accurate, but details may have changed since we last looked. If you’ve noticed an error, please email us.
Scores inform. Enforcement decides.
Luna answers “how good is this output?” quickly enough to ask on every request. in/guard/out asks a different question – “does this ship?” – and is built to act on the answer: deterministic grounding catches the numeric class outright, claim audit ties assertions to sources, FIX mode repairs what has a safe fix, and the agent stack extends the same authority to tool calls and budgets. If your team already scores with Luna, nothing conflicts: the wire enforces while the platform trends.
Frequently asked questions
Luna is fast and cheap – isn’t that enough for guardrailing?
Fast scoring solves half the problem. A score is a judgement, and a threshold on it trades misses against false alarms. in/guard/out starts with checks that cost nothing and need no threshold – a figure either matches its source or it doesn’t – and its verdicts end in repair where a fix exists, not only a block.
What changes now that Galileo is part of Splunk?
For Splunk shops, Galileo becomes the natural place to observe and evaluate agents next to the rest of their telemetry. in/guard/out is unaffected by the choice: it is a vendor-neutral enforcement point in front of any model provider, and it composes with Splunk Agent Observability the way it did with Galileo.
Do the platforms conflict?
No. Galileo instruments your app for evaluation; in/guard/out sits on the wire for enforcement. Production catches from the proxy make strong labeled cases for whatever evaluation platform you trend quality in.
What does enforcement add for agents specifically?
The failures that hurt are actions, not sentences: the ungrounded payee, the loop that burns budget, screened PII exiting via a tool argument. Those need run-scoped enforcement – budgets, grounding, taint, ordering – decided on the run’s accumulated state, which a per-action score doesn’t carry.
See it on your traffic first.
FIX mode observes without blocking, so you can measure what in/guard/out catches before enforcing anything. We are running a limited demo - sign up and we will get you in as soon as we can.