Connect the estate your AI already runs on.
Strai8 reads from the systems you already have — cloud accounts, document libraries, source control, observability, security and identity — to find the AI in your estate and keep its evidence current. Every connection is read-only.
Cloud providers
Managed AI services run here, often without anyone raising a ticket.
So OpenAI and ML deployments are inventoried rather than assumed.
Vertex endpoints and notebooks are production AI the registry should hold.
Document libraries
Your policies and assessments are the evidence every framework asks for.
Governance documents are only useful to an audit if they can be found.
Runbooks and decision pages carry controls that never reached a policy.
Wikis hold the AI decisions nobody wrote up formally.
Signed assessments and vendor reports usually live in a shared folder.
Source control
Most AI arrives as a dependency or an SDK call, not a purchase.
Self-hosted repositories hide exactly the same imports as cloud ones.
A model called from code is a system, whoever owns the repo.
Observability
Logs show which models actually ran, not which were declared.
Runtime traffic proves a system is live and shows who it serves.
Metrics tell you when a model's behaviour moved, and when.
The events you already collect name AI activity nobody reported.
Telemetry ties a running service back to an owned AI system.
Security
Endpoint telemetry gathered for security also names the AI tools in use.
Egress logs reveal which AI services your network genuinely reaches.
Identity providers
A detection is only actionable once it resolves to a person.
App assignments show who was granted which AI tool, and when.
Groups and departments turn a hostname into an accountable owner.
Model providers & gateways
Direct API usage is AI spend and AI risk that no inventory sees.
Workspace usage shows which teams built on a model, and how much.
Deployments made inside a subscription rarely reach a governance review.
Foundation-model calls are third-party AI under your own account.
Model monitoring & evaluation
Drift and quality metrics are the evidence a reliability control wants.
Explainability output is what an interpretability obligation asks to see.
Data health tells you when a model's inputs moved underneath it.
Traces and evaluations are the test record for an LLM feature.
MLOps & model registries
The registry already knows a model's version, owner and lineage.
Downloaded models carry licence terms you inherit on use.
An endpoint is production AI whether or not anyone registered it.
CI/CD
Pipelines pull models and credentials at build time.
Workflows call AI services on every push, unreviewed.
Container registries
Images bake in frameworks and model weights you then ship.
A pushed image is the most reliable fingerprint a workload has.
Scanning images finds AI dependencies before they reach production.
Collaboration
Assistants approved by nobody show up as installed apps first.
Bots and copilots reach company data through the chat client.
And many more
New connectors ship continuously, in the order customers ask for them. Tell us which one you need.
Need a custom integration?
Tell us what your stack runs on. If it holds AI, holds evidence or names an owner, we can connect it — and connectors are built in the order customers ask for them.