nexus-totvs-platform

You can turn TOTVS sales transcripts into action plans by running a fixed five-tool Pydantic AI agent workflow. This pipeline structures raw meeting data into audit-friendly outputs, checking the product catalog and creating development plans to prevent churn risks from getting lost.

SSE streaming works with Pydantic AI by streaming tool calls, results, usage, and completion events from the Python agent service through a Node Fastify proxy to the frontend. This end-to-end streaming provides live visibility into the agent pipeline execution.

Yes, you can use multi-provider model fallbacks for TOTVS transcript processing through FallbackModel support. This accepts provider shorthand and implements failover to reduce single-model downtime during the agent workflow execution.

Yes, the pipeline applies PII redaction for TOTVS inputs using pre-flight detection for already-anonymized tokens. It also blocks prompt injection attempts via a prompt guard to ensure dataset-safe security before analysis.

Yes, you need canonical NEXUS identifiers to process TOTVS transcripts through the agent workflow. The pipeline requires these identifiers to properly execute opportunity updates and route development plans within the NEXUS dataset.

The best way to persist TOTVS sales transcript analysis results into PostgreSQL is through the pipeline's database-integrated audit trail. It automatically saves analysis results, identified gaps, development plans, and team notifications using the project's canonical schema.