POST /chat/intelligent
Streaming inference through the private control plane. Use model: auto when routing is enabled.
Developer docs
Implementation notes for private AI infrastructure.
A practical first-pass reference for teams evaluating Gridlight: API shape, deployment patterns, security posture, and what to validate during a technical briefing.
Unified API
One private endpoint for every AI workload.
Applications should not hardcode model vendors or hardware paths. They call Gridlight, and the control plane handles model availability, policy, capacity, routing, and lineage.
const response = await fetch('https://your-gridlight-instance/chat/intelligent', {
method: 'POST',
headers: {
Authorization: `Bearer ${token}`,
'Content-Type': 'application/json'
},
body: JSON.stringify({
model: 'auto',
policy: 'private-default',
messages: [{ role: 'user', content: prompt }],
stream: true
})
});POST
POST /neon
Synchronous JSON response for application workflows that do not need streaming.
GET /models
List available models, versions, classifications, and routing policies.
GET /capacity
Read TFLOP allocation, queue depth, and worker-node telemetry.
GET /audit/logs
Export inference lineage and policy decisions to your SIEM or audit store.
PUT /policy
Update governed routing, data classification, and access policies.
Security posture
Privacy by architecture, not by checkbox.
Data gravity
Inference follows the data instead of moving data to a shared endpoint.
Cloud AI patternexternal transmission events
Your dataCloud endpointApp
- ✕ Data leaves the customer environment per request.
- ✕ External vector DB or vendor logs may expand review scope.
- ✕ Data movement creates latency and vendor risk review work.
Gridlight-controlled patterninside configured boundary
Local dataLocal RAGLocal modelLineage
- ✓ Local retrieval and context assembly.
- ✓ No external vector database required.
- ✓ Full lineage recorded in the customer audit path.
Security posture
- Data, prompts, and model outputs stay inside your environment by default.
- Your identity provider governs access; Gridlight should not become a parallel identity silo.
- Audit records include user, timestamp, model, policy decision, context lineage, and hardware path.
- Compliance positioning is architecture-friendly; Compliant by Design remains roadmap until attestations are confirmed.
Deployment topologies
Deploy where your work and data already live.
Single on-prem cluster
Start with one data-center deployment for internal apps, local RAG, and governed inference.
Federated edge sites
Run inference close to plants, clinics, branches, or robots while maintaining centralized policy.
Private cloud / sovereign cloud
Use the same API and governance model inside controlled private-cloud environments.
Cloud fallback if configured
Keep application architecture stable while selectively allowing cloud AI as a fallback or preferred route.
Technical briefing