1What it is
BOTY is one instance-aware conversational engine with two independent interfaces:
Full-screen application
A complete chat workspace behind login: sidebar, prompt manager, voice in and out, file upload, streaming answers.
Website support widget
A lightweight chat button that drops onto any web page, including mobile web, with the same grounded answers.
Neither interface contains its own AI logic. Fix or improve the engine once and both benefit. Configuration resolves per instance at run time, so one codebase serves every assistant we create (internal, customer-facing, per department). When a question needs a human, the Talk to a Person add-on hands the conversation to our staff (see C5).
2Capabilities
| Streaming | Word-by-word answers, with automatic fallback for remote users |
| Prompt manager | Suggested (one-click), smart (interview-style templates) and saved prompts |
| History & summarization | Conversations stored in the database; long chats auto-summarized to bound cost and latency |
| Guardrails | Per-instance safety and scope constraints, enforced before the model answers; strict "retrieved data only" mode |
| Personas | Answers framed for the audience (executive, technical, customer) |
| Input / output | Text, voice capture, file upload; streamed text and spoken TTS; export to PDF / Word |
| Citations | Sources surfaced with grounded answers (toggle per instance) |
| Agent tools | The bot can take registered actions, not only answer |
| DEV view | For administrators: switch model, mode, knowledge group and clearance live; prompt inspector and debug payload show exactly what was sent to the model |
3Scores
| Category | Score |
|---|---|
| Architecture & design | 9.0 |
| Multi-interface delivery | 9.0 |
| Prompt system | 9.0 |
| RAG grounding | 9.0 |
| Streaming & real-time UX | 8.8 |
| Conversation management | 8.8 |
| Observability & debug | 8.8 |
| Code quality & modularity | 8.7 |
| Security & guardrails | 8.5 |
| Reliability & performance | 8.5 |
4Objections the review answers
| Objection | Answer in the review |
|---|---|
| "We already had a chatbot. It was not worth much." | It was probably answering from model training data or a keyword FAQ. The difference here is a citation on every answer. |
| "Different departments need different answers." | One knowledge foundation, many instances, each with its own persona, scope, guardrails and look. Each instance gets its own database (36 tables) in one operation. |
| "What stops it saying something we would not want?" | Per-instance guardrails, strict mode, and every conversation retained for audit. |
| "Streaming breaks behind reverse proxies." | Server-sent events with proxy buffering disabled; documented as a required installation step. |
5What our team should probe
| Probe | Our finding |
|---|---|
| Streaming through our reverse proxy / load balancer and network | [p_c3_1] |
| Login integration with our identity provider for the full-screen app | [p_c3_2] |
| Human handoff route for customer use: who receives escalations? | [p_c3_3] |
| Conversation retention and privacy notice for customer-facing use | [p_c3_4] |
| Branding, accessibility and language requirements | [p_c3_5] |
6Verdict, as stated by the review
Approved for production, grade A-. A unified, instance-aware engine that drives a full-screen application and an embeddable website widget, with streaming, persistent history, summarization, orchestration, guardrails, personas, voice and agent tools.
Our assessment: [as_c3]. [as_c3_why]