Nexa IntelligencePLATFORM
Install it in your cloud. Run it as a business unit.
Nexa Intelligence is not a tool you log into. It is a platform you operate. You request the Cloud Harness, we install it inside your own Cloudflare, AWS, Azure, or on-premise environment, and from that moment the three stages run inside your perimeter. Workers and customers are interviewed by our AI conversation agent at three cents per enriched conversation. Each conversation lands in your tenant as ten typed dimensions of structured intelligence, refreshed on a weekly cadence. Your dashboards, your brand-partner API, and your own AI agents all read from the same corpus, with citation back to a real human interview.
This section of the site is about how you roll the platform out, how you operate the loop, and how you, as an individual operator, use what the platform produces. The mechanisms underneath are documented in two places. The interview protocol, the enrichment pipeline, and the ten-dimension schema live under Methodology. The data path, the tenant model, and the refusal protocol live under Trust & Security. Where this section would otherwise duplicate either, it points across rather than repeating.
CLOUD HARNESS
The install that turns your cloud into the platform.
The Cloud Harness is the deployment artifact. You request it, we provision it inside your own cloud account, and it brings the entire platform with it: the voice capture pipeline, the enrichment workers, the dual-store fact corpus (graph for discovery, Postgres for pivots), the MCP tool surface that brand-partner APIs and AI agents query through, and the operator console where your team configures the interview agent and reviews execution. You own the install the moment it lands.
Once the harness is up, the platform runs inside your perimeter. Audio enters under our zero-retention contract with ElevenLabs, gets stripped of personal identifiers in-process by an in-memory worker, and is rendered back as a synthetic voice. The original recording is never written to disk. Every enrichment call runs against a system prompt registered by SHA-256 hash, against your data, inside your tenant. If you end the vendor relationship with us, the harness keeps running.
The harness ships with every integration the platform needs: the LLM contract with Anthropic (no training on every plan, no retention on Enterprise), the voice contract with ElevenLabs (no retention on every plan), the storage and compute on the substrate you chose, and the operator MCP that lets your team add interview tracks, edit the interview agent’s protocol, extend the redaction list, and read the per-conversation audit. Setup is a guided process. You do not have to staff a platform team to operate it.
HOW THE LOOP RUNS
The Intelligence Layer you operate
What you actually get once the harness is installed: a per-tenant corpus where every fact is timestamped, segment-scoped, and cited back to a real interview. The shape of the data, the three consumption surfaces, and how your team queries it.
Read more →Stage 01. Gather
The product view of how you launch interviews. The AI interview agent runs the conversation over voice or cellphone, fifteen to forty-five minutes per worker or customer, at three cents per enriched conversation. You configure the tiers, define the cohort, and watch the corpus fill.
Read more →Stage 02. Codify
The product view of how raw conversations become a queryable layer. Ten typed dimensions per vertical, bi-temporal validity per fact, weekly refresh by default. What you see in the console while the pipeline runs, and what lands in the corpus when it finishes.
Read more →Stage 03. Monetize
The product view of the three endpoints that consume the corpus: your internal dashboard, your brand-partner API, and your AI-agent retrieval feed. How you turn each one on, how you scope access, and how every answer carries citation back to a chunk.
Read more →ARCHITECTURE AND OPERATION
Architecture
The operational view of the dual-store: a graph for discovery, Postgres for pivots, joined at the chunk-ID spine. MCP is the contract surface every consumer reads through. The deep engineering reference ships with the Cloud Harness; this page is what you need to operate the platform.
Read more →Security in the platform
A short operational summary of the security model: in-tenant deployment, per-tenant scoping in the data access path, zero-retention on voice, no-training on LLM. The full posture, the refusal protocol, and the subprocessor list live under Trust & Security.
Read more →DEEPER READING
Methodology, in detail
The interview protocol (anchor-gated probing, neutrality enforcement, wrap protocol), the five-call enrichment pipeline, the deterministic scoring formula, and the ten dimensions per vertical. Where Stage 01 and Stage 02 came from.
Read more →Trust & Security, in detail
The mechanism behind every confidentiality and security claim: synthetic voice substitution, the verbatim provenance chain, the per-conversation interviewer audit, the security cascade in the data access path, and the things we refuse to support.
Read more →