The building-block catalog

33 blocks across 7 tiers, researched against what real chatbot platforms and current papers actually ship in 2026 — not guessed. Status is marked honestly: what's live in this demo versus what's designed and coming next.

Live6 tiers

Partial1 tier

Planned0 tiers

Foundation

The core pipeline every chatbot on this platform starts with.

LiveFully wired in this demo (sentiment is upgraded by the Jev decision model through OpenRouter)

Model provider & routing (OpenRouter)
RAG / knowledge retrieval (pgvector)
Long-term memory (Mem0)
Sentiment detection & empathy routing
Tool / product recommender
Ticketing & complaint tracking
Rate limiting, CORS, analytics

Agentic & Reasoning

What separates a chatbot from an agent — the 2026 differentiator tier.

LiveAll 7 blocks in this tier are wired

Reasoning-mode toggle (extended thinking)
Self-reflection / critique pass
MCP tools (real Model Context Protocol server)
Multi-agent handoff (router → specialists)
Confidence-gated routing + semantic caching
Hybrid retrieval + reranking (cross-encoder)
Tool-call verification & fallback/retry

Trust & Safety

Defense-in-depth, not a hack-proof claim — see the research notes.

LivePII redaction, AI-disclosure, human handoff, prompt-injection guardrails, and policy-adherence checking are all wired

Input/output guardrails, prompt-injection detection
Real-time PII redaction (Presidio)
Business-policy adherence guardrail
AI-disclosure banner
Human handoff / escalation queue

Proactive & Personalization

Reaching out before the user has to ask.

LiveAll 3 blocks in this tier are wired

Behavioral trigger engine (idle / scroll)
Hyper-personalization via user-state modeling
Multi-language auto-detection

Observability

Resolution rate, not deflection rate, is the metric that matters.

LiveFully wired: visitors are asked whether their problem was solved, the owner gets a private insights page, and any bot can run a real A/B test

LLM-as-judge auto-evaluation
Resolution-rate analytics
A/B testing of prompts / models

Channels

Where the chatbot actually shows up.

PartialWeb chat is live; other channels are guide-only

Web widget
WhatsApp (guide-only)
Slack (guide-only)
SMS (guide-only)
Voice (guide-only)

Built-in Evaluation Report

A real, on-demand RAGAS scoring report for any build — no other builder ships this inline.

LiveOn demand from Studio's "done" screen or any /bot/[id] page ("Run evaluation report"); it needs the Python evaluation service deployed (ML_SERVICE_URL) and says so plainly instead of faking a score when it is not

Real 2-turn test conversations (FAQ question + a fixed follow-up)
RAGAS scoring: faithfulness, answer relevancy, context precision
Powered by RAGAS (Python microservice)

Configure it yourself in the Studio

Advanced Mode lets you toggle 21 wireable blocks per bot — RAG, memory, sentiment, agentic tool use, guardrails, and more — and see the effect live in the preview.

Open the Studio