Health status & pending recommendations

A health score monitors your LLM APIs. Anything it finds arrives as a pending recommendation, and nothing runs until you approve it.

The guardian's four moves (illustrative story)

90

Detection“helpfulness” quality is slipping
Quality
95
Errors
97
Safety
100
Follow-up
66

A four-part health score

Quality, errors, safety, and follow-up, each scored 0–100 with daily history. Errors and follow-up are counted from records; quality and safety are model-judged with evidence.

Example
Quality 95 /100
Errors 97 /100
Safety 100 /100
Response 66 /100

Overall score, last 14 days

Pending recommendations

Findings arrive as proposals with approve / dismiss actions and step-by-step fixes — for wasted cost, with a cheaper model to switch to and the estimated savings. Nothing runs until you approve.

Example
Pending proposals3
CostModel switch proposal
est. savings $12/moApprove Reject
QualityFix for “conciseness” drop
with apply stepsApprove Reject
SafetySafety hardening proposal
with apply stepsApprove Reject

Findings notifications

The moment a new proposal lands, a notification carries the gist: what was found and how much it helps. Open the proposal straight from the message and approve or reject it there. Email on Free; Slack, webhooks and every channel on Pro and above.

Example

ai-alerts

Argosvix APPjust now

The guardian found a new proposal

CostModel switch proposalest. savings $12/mo, with apply steps

Open proposal

Delivered to: email (from Free) / Slack, webhooks and every channel (Pro and above)