Context-aware answers
Questions can include device IDs and alert IDs so the copilot answers inside the current operational context instead of replying generically.
AI Operations
Copilot assistance with guardrails, sources, and tenant isolation
47Dynamics exposes AI as an operational acceleration layer, not an unbounded autonomous actor. The copilot combines tenant-scoped retrieval, streamed answers with sources, rate limits, and human-approved follow-through so operators move faster without surrendering control.
What the copilot does in production
Streamed responses with sources
The service supports token streaming and returns cited sources first, so operators can review where the answer came from before acting on it.
Suggested actions, not blind autonomy
Copilot answers can include structured suggested actions such as runbook execution, ticket creation, or guided navigation, but the operator remains in control of what is actually executed.
Governance and safety model
Tenant-scoped retrieval
The retrieval layer scopes ChromaDB collections per tenant when available, isolating AI knowledge context and preventing cross-tenant document leakage in the same shared service.
Prompt-injection defenses
The AI service includes dedicated test coverage for prompt injection and role-switch attempts, with sanitization applied on user-facing inputs across the retrieval and answer path.
Per-tenant throttling
Copilot requests pass through a tenant rate limiter before the model is invoked, preventing abusive usage patterns and protecting the service during concurrent incident peaks.
Bounded request surface
Requests are strongly typed and bounded: tenant ID required, question length capped, and context arrays limited so AI remains an operational tool with explicit, reviewable inputs.
Deployment model
FastAPI service
AI operations run as a dedicated Python service exposed behind the platform gateway, with health checks, structured API contracts, and explicit operational boundaries.
RAG over ChromaDB
Retrieval-augmented answers use ChromaDB-backed vector search so the copilot can cite platform knowledge, runbooks, and tenant-specific operational context instead of hallucinating from a blank prompt.
Flexible model backend
The service is designed for OpenAI-compatible APIs or local models, including GPU-backed deployments when throughput or data-residency requirements make local inference the right choice.
Where AI creates operational leverage
Incident triage
Operators can ask what changed on a device, what policy is implicated, and what first-response steps are safe before escalation, without leaving the active incident context.
Change planning
Before patch windows or large tenant rollouts, the copilot can summarize runbooks, prerequisites, and likely blast radius so teams review faster and execute with fewer surprises.
Audit and evidence prep
The copilot helps operators assemble the right reports, logs, and evidence paths for customer reviews or compliance cycles, reducing manual hunting across tools.
Assess AI operations like any other controlled platform capability
We recommend reviewing AI on four axes: source quality, tenant isolation, operator control, and measurable time saved. We can walk through your runbooks, incident patterns, and evidence requirements in a structured demo instead of relying on generic AI claims.
