One conversation. Every answer.

EVO Products' Cloud 9 portal handled orders, invoices, inventory, support, and reporting for their distributor customer base. But power came at a cost: getting to any piece of data required users to know exactly where to look; multiple screens, filtered views, downloaded reports, one task at a time.

For users who needed to check order status, pull a historical invoice, or verify IPP stock, the process was identical regardless of how routine the question. Navigate. Filter. Wait. Download. Repeat.

• Fragmented navigation for routine tasks: A question like "what's the status of my last order?" required navigating to a specific screen, applying filters, and reading tabular data — every single time.

• Historical data was painful to access: Invoice history, transaction records, and prior quotes were buried behind date filters and pagination. Users had to work through date filters and pagination to locate historical records — an inefficient process for what should be routine lookups.

• No unified interface: Orders, quotes, inventory, support cases, and reports each lived in separate sections with no shared context. Every task reset the user's journey from scratch.

• Routine tasks still required manual navigation: Questions that could be self-served required navigating the same complex portal flows — adding friction to every routine inquiry.
  

Cogent Labs built STAC-E — a conversational AI agent embedded directly into the Cloud 9 portal. Users ask questions in plain English. STAC-E handles retrieval, filtering, aggregation, and document generation on their behalf — scoped strictly to their authenticated account.

• Natural language replaces UI navigation: Users ask questions the way they think them. STAC-E interprets intent, selects the correct data source, and returns the right answer — no menus, no filters, no navigation overhead.

• Document retrieval on demand: Invoices, quotes, price lists, and reports are returned as downloadable PDFs or CSVs in the same conversation thread. Single files or bulk exports, handled in one request.

• Multi-record queries in a single turn: "Show me all open orders from last month" or "get Q1 invoices" retrieves, paginates, and presents multiple records without requiring multiple requests or page visits.

• Support case management built in: Users can view open cases, read case history, and add comments directly from the chat — eliminating the round-trip to the support section entirely.

• Knowledge base powered by real content: How-to questions and product queries are answered from indexed PDFs, tutorials, and the Bookstack knowledge base — not generic AI responses. Every answer is grounded in EVO Products' own content — not general AI training data.

• Continuous improvement through fine-tuning: User corrections and evaluation data feed back into model improvement cycles — making STAC-E more accurate over time without requiring a full model retrain from scratch.

STAC-E exposes 12 distinct capability domains through a single conversational interface, each backed by secure, account-scoped tool execution via the MCP server/client architecture.

Orders & Commerce

1. Sales orders: view, filter, sort, paginate, download
2. Invoices & quotes: search, view, bulk PDF export
3. Transaction history: payments, deposits, records
4. Multi-record queries in a single request

Inventory & Samples

1. Sample request tracking and status
2. IPP inventory availability lookup
3. Inventory release request submission
4. Product and item search with detail view

Support & Cases

1. Open support case visibility
2. Case history and timeline access
3. Add comments directly from chat

Knowledge & Intelligence

1. Knowledge base: PDFs, websites, tutorials
2. Reports and price lists: PDF/CSV download
3. Continuous learning via user corrections
   

The STAC-E stack combines a frontier AI model with a structured tool-execution layer — MCP — giving every conversation access to live portal data, while Logfire captures every call for traceability and debugging.

STAC-E transforms the Cloud 9 portal from a navigation problem into a conversation — and the downstream effects show up in support volume, task completion time, and user confidence with the platform.

• One Interface: Orders, invoices, inventory, support, and knowledge base accessible through a single conversational entry point — no multi-screen navigation.

• 0 Clicks: Tasks that previously required multiple screens and filter steps are reduced to a single natural-language request with an immediate, formatted response.

• Bulk Exports: Users request multi-document downloads — invoice bundles, report sets, price lists — without manually selecting and downloading each file individually.

• Self-Serve: Customers view case history, get status updates, and add comments directly in chat handling routine queries without navigating to the support section.
  
  

The build: system architecture on screen, code in editor, the full stack coming together.

The Cloud 9 platform wasn't broken — it was comprehensive. The problem was that comprehensiveness had a cost: users had to carry the cognitive load of knowing where everything lived, how to filter it, and how to export it. STAC-E eliminates that load entirely.

The solution Cogent built treats the portal's full data surface as a set of secure, account-scoped tools that a conversational agent can call on demand. Users speak their intent in plain language. STAC-E determines which tool to call, calls it within the user's account boundary, and returns a formatted, usable response — immediately, in the same conversation.

• Claude Sonnet 4.5 handles intent interpretation and response generation with a structured prompt system tuned for EVO Products data patterns

• The MCP server/client layer provides the secure, structured bridge between conversational intent and live portal API calls

• Book stack knowledge powers tutorial and product FAQ answers without hallucination — responses are sourced, not generated from general training data

• Log-fire traces every tool invocation, making debugging and performance monitoring production-ready from day one

• The continuous fine-tuning pipeline captures user corrections to improve accuracy iteratively, without requiring a full model retrain

The result is a platform where the correct answer to any customer question is one sentence away — and where the system gets more accurate over time, not less.

Cogent Labs designs and builds AI agents that connect to your existing systems, respect your security model, and reduce operational overhead — without replacing the infrastructure you already run.