The operational intelligence layer for AI agents

Know what your agents are doing, why things break, and where to make them better. Observability and intelligence from one instrumentation, grounded in your real systems.

Book a demoRead the docs
Invariance
Observe
Runs
Signals
Reviews
Monitors
Agents
Evals
Intelligence
Impact
Divergence
Graph
Ask
Knowledge
Admin
Billing
Settings
RunsResearch pipelinerun-2847
Research pipelinerunningresearch-agentnodes0·tok3,110·cost$0.042·elapsed3.5s
DRoute reque..orchestratorTweb_searchresearch-agentTparse_resul..research-agentSsummarizesummarizer-sub!send_emailresearch-agent
Trace Steps
1DRoute request42ms
2Tweb_search1.2s
3Tparse_results89ms
4Ssummarize2.1s
5!send_email87%67ms
Node Detail
Select a node to inspect
Narrative

Orchestrator routed query to research-agent, which delegated search and parsing to sub-agents. Unauthorized send_email flagged at 87% anomaly. Budget check passed within limits.

Key Moments
Sub-agent spawned for summarization
send_email flagged — not in allowlist
Budget verified: $4.20 / $5.00

Observability that watches itself.

Catch drift, missed steps, policy violations, and runaway spend the second they happen. Monitor agents evaluate every run for you, around the clock.

Findings arrive ranked by severity, with the rule that fired and the evidence already attached. One click to escalate, ack, or send to human review.

tap to pause
Monitor Agent
policy:refund_v2 · drift_threshold · pii_scan
watching 0 events
refund-bot
support-triage
deploy-bot
tap to pause
run_8421 reconciledrefund-bot0 sources linked
Tool calls
fetch_customer({id:"acct_8821"})
lookup_policy("refund_v2")
issue_refund({amount: 420.00})
check_approval("ch_3KqZ5L")
slack.post("#refunds-out", "...")
Linked entities
Salesforce
Acme Corp
Enterprise · LTV $84,210
N
Notion
Refund Policy v2
approval required > $200
S
Stripe
Refund · $420.00
ch_3KqZ5L · succeeded
Approval LogDIVERGENT
approval not found
policy violation
#
Slack
#refunds-out
delivered · 2 reactions

Every run, grounded in your real systems.

Connect Stripe, Salesforce, your CRM, your knowledge base, your approval logs. Every tool call links to the real-world entity it touched, automatically.

Stop trusting agent self-reports. Reconcile what the agent says it did against what your systems actually show, and catch silent failures the moment they appear.

Stop digging. We tell you what to fix.

Invariance learns your operations and surfaces the highest- impact fixes automatically. Each suggestion arrives with the affected runs, the divergent step, the baseline, and the dollar impact already attached.

One click to enable the proposed monitor. One click to compare against a known-good run. Less dashboard archaeology, more shipping.

tap to pause
Routed suggestionsanalyst agent · auto-surfaced
HIGHmissing approval
Refunds bypassing manager approval
refund-bot
MEDIUMstale policy
support-triage citing policy v1.3 (deprecated)
support-triage
HIGHout of order
deploy-bot skipping smoke test step
deploy-bot
IMPACT
$4,820
12 events · last 7d
runs affected12
fidelity78%
baselinerun_7102
Proposed monitor
refund_amount > 200 && !has_evidence('approval')

Built for teams shipping agents to production.

One SDK. Ten minutes to first trace. We’ll walk you through it on a 20-minute call.

Book a demoSee pricing