See every AI call in your cloud in 10 minutes
Auto-discover AI agents, LLM egress, and ownership gaps across your cloud — with read-only eBPF sensors.
Inventory today, lineage and control when you need it.
TTFF ≤ 10 min
No content capture
<2% CPU • <50 MB
Live Detection
Runtime inventory
Find every service calling an LLM
We fingerprint outbound TLS + process hints to attribute LLM usage to real services, namespaces, and IAM roles.
AI Runtime Inventory
6 services detected
| Service | Namespace | SA / IAM Role | Exposure | Endpoint | Calls (7d) | Bytes (7d) | Last seen | Owner (lite) | Egress | Databases |
|---|---|---|---|---|---|---|---|---|---|---|
| invoice-ai | prod-finance | sa-invoice-ai | external | OpenAI | 4,812 | 1.2 GB | 2m ago | finance-ml | third-party | PSQL,MySQL |
| webapp-next | prod-app | sa-frontend | external | Anthropic | 3,821 | 860 MB | 6m ago | unknown | third-party | Oracle(x2) |
| auth-mgr | dev-ai | sa-recon | internal | Vertex AI | 1,210 | 340 MB | 13m ago | platform-ai | private/VPC | PSQL (x3) |
| content-moderator | prod-safety | sa-moderation | internal | OpenAI | 2,156 | 520 MB | 18m ago | trust-safety | third-party | MongoDB |
| recommendation-api | prod-ml | sa-ml-service | external | Bedrock | 6,240 | 2.8 GB | 5m ago | ml-platform | third-party | Redis,PSQL |
| data-pipeline | staging-analytics | sa-etl-worker | internal | Cohere | 892 | 180 MB | 31m ago | unknown | third-party | Snowflake |
← Scroll horizontally to see all columns →
Owner (lite) is inferred from labels/tags
Confidence blends endpoint + SNI + process hints
Privacy: metadata only — no prompts or outputs
Watch AI Observability in Action
Our eBPF sensors detect AI services in real-time - at lightning speed
eBPF Sensor Active
0 Eliminating
Shadow AI detection
Your engineers are shipping AI features faster than you can track them.
Flag services making LLM calls without owner or review tags, first-seen behavior in prod, or wasting your resources
Prod service 'webapp-nextjs' calling OpenAI
High- No owner tags found
- Interacts with 2 medium sensitivity apps
- Egress volume much higher than your baseline.
Last seen 6m ago
'auth-mgr' reading user tables + calling Vertex
Medium- No security_reviewed tag
- Accesses 4 low risk DBs with PII
- Accesses 1 DB identified to hold user data
Last seen 22m ago
How it works
Read-only sensors. Real-time results.
Node eBPF collectors observe outbound connects and process hints. A curated endpoint catalog attributes LLM usage with confidence scoring.
Deploy
Helm one-liner installs DaemonSet on your cluster.
Observe
Traffic on your cluster + process hints.
Attribute
To providers like OpenAI, Anthropic, Bedrock and many more.
Flag
Shadow AI: no owner/review, newly seen, unusually high volumes
Metrics
Get comprehensive visibility into your AI usage
Track AI calls, detect patterns, and monitor your infrastructure in real-time.
AI Calls (last 14d)
demo data
Agents detected
37
Active services
LLM providers
5
OpenAI, Anthropic, Bedrock, Vertex, Custom
Shadow AI flags
6
Requires attention
Privacy & performance
Metadata-only by design
We never collect prompts, responses, or secrets. Sensors are read-only and bounded.
Sensitive Content captured
None
Metadata only
CPU overhead
< 2%
p95 < 5%
Fast
< 10 mins
MTTV
Add-on value
Upgrade to AI Data Guard when you need lineage and control
Keep the free runtime inventory. Add sensitive→LLM lineage, intent policies, and regulated alerts when you're ready.
Free — AI Runtime Inventory
- Auto-discover AI agents and LLM endpoints
- Shadow AI flags (no owner/review, first-seen in prod)
- Owner (lite) inference from labels/tags
- Database fingerprints (Oracle, PSQL, MySQL, Redis and more)
- Data classification and sensitivity labels (lite)
- Standard Login
- 7-day lookback, community support
Paid v1 — AI Data Guard
- Sensitive→LLM lineage (Snowflake, Postgres, Redshift, Athena)
- Intent policies by team/service and provider allowlists
- Regulated egress alerts with evidence
- SIEM export • 30-90 day retention
- SSO support
- Deep Data Classification and Sensitive data scans
- 24/7 Support
Install
Get signal in minutes
Use Helm to install in a jiffy. Don't like it ? Uninstall is one command.
One-Line Installation
AGENTIC INSTALL WIZARD
curl -fsSL https://raw.githubusercontent.com/aurva-io/AIOstack/main/install.sh | bash
✅ Checks prerequisites • ✅ Guides credential entry • ✅ Installs & verifies • ✅ ~2 minutes total
Uninstall is just as easy
UNINSTALLER
curl -fsSL https://raw.githubusercontent.com/aurva-io/AIOstack/main/uninstall.sh | bash
FAQ
Common questions
Do you collect prompts, responses, or secrets?
Never. We're metadata-only by design. To elucidate, we see that you called OpenAI, not what you sent.
What environments are supported?
Kubernetes(EKS/GKE) is the primary path. Feel free to reach out to us for your needs.
Can I uninstall easily?
Yes. One Helm uninstall detaches all kernel hooks. No residual processes remain. Although do note that we will be sad to see you go :(
How is this different from Wiz or Datadog?
Wiz does cloud security posture. Datadog does observability. Neither shows AI-specific visibility like PII exposure in LLM calls or Shadow AI detection.They
I am a large finacial org and need a PaaS option.
While we encourage SaaS, we'd be happy to help you out with a PaaS deployment if you have the genuine need. Reach out to our team for more info :)
I read through all this, but didn't understand what AIOStack does.
No problem. AIOStack is an eBPF based tool that you can deploy to Kubernetes to discover, track and monitor AI services no matter what they are named or claim to be. Additionally we can tell you what databases, applications and external domains they communicate with and if you should be concerned about its behaviour.
I really liked the product and want alerts, SSO and integration with Slack and Jira.
We are thrilled that you liked our product ! Reach out to us so that we can upgrade you to the Data Guard plan.
I see a lot of "AI" here, will the Data Guard plan cost me a bomb ?
Haha, not really. We know that pricing can be scary, reach out to us and expect to be pleasantly surprised by how much we can save you :)