Skip to content

Hybrid Chat Agents

A 4-tier resolution cascade — deterministic rules, knowledge base retrieval, LLM fallback, and human escalation — in a single agent. Works across LINE, Instagram, WhatsApp, Facebook, and YouTube.

Resolution Cascade

Four tiers. One agent. Every message resolved.

Incoming messages flow through each tier in order. The first tier that matches with sufficient confidence responds. If no tier matches, the message is escalated to a human.

Incoming customer message
1

Rule Table

Embedding-based cosine similarity against CSV-imported rules. Deterministic, fast, zero LLM cost.

Fastest
No match or low confidence
2

Knowledge Base / RAG

Semantic retrieval from uploaded documents. Configurable minimum similarity threshold.

Grounded
No match or low confidence
3

LLM Fallback

Full AI generation with system prompt and conversation context. Uses your BYOM provider.

Creative
No match or low confidence
4

Human Escalation

Hand off to human when confidence is low. Thread marked as escalated.

Safety net

Messages cascade through each tier. The first tier that matches with sufficient confidence handles the response automatically.

Capabilities

Everything you need to build intelligent chat agents

From rule import to analytics, every tool you need to create, test, and monitor your resolution cascade.

CSV Rule Import

Upload spreadsheets of Q&A pairs. Patterns are auto-embedded for paraphrase matching. No code required.

Embedding-Based Matching

Cosine similarity against rule centroids handles paraphrases, typos, and variations. Not brittle keyword matching.

Multi-Channel

Same agent works on LINE, Instagram, WhatsApp, Facebook Messenger, and YouTube. One configuration, every channel.

Conversation Threading

Multi-turn context preserved across messages. Automatic thread compaction summarizes older messages.

Interactive Testing

Test your agent's resolution cascade in a chat-like UI. See which tier responded and the confidence score.

Analytics Dashboard

Source distribution, daily trends, top rules, unused rule detection, and channel breakdown.

Quick Create

From zero to production in under 2 minutes

Three fields — name, description, channel — and your agent is live. Smart defaults resolve your BYOK provider and model automatically. Choose an industry template for a head start.

Quick-create flow

1

Name & describe

Give your agent a name and one-sentence description. That's the minimum.

2

Pick a channel

Choose LINE, Instagram, WhatsApp, Messenger, or any of 12 supported channels.

3

Auto-configure

Smart defaults resolve your BYOK keys to the best available provider and model. Governance layers activate automatically.

Governance Score included

Every agent displays a real-time 0-100 governance score based on 8 weighted factors — escalation policy, audit logging, knowledge base, rules, RBAC, cost controls, memory limits, and BYOK encryption.

10 industry templates

E-Commerce Support
Healthcare FAQ
SaaS Onboarding
Restaurant Booking
Real Estate
Travel
Education
Finance
Legal
HR

Each template includes a system prompt, suggested rules, knowledge base structure, and escalation policy. Fully customizable after creation.

Use Cases

Real-world resolution cascades

See how the 4-tier cascade handles different industries and message types.

Healthcare Clinic FAQ Bot

Import 200+ FAQ rules from a spreadsheet. Rule tier handles routine questions instantly. RAG tier pulls from clinical guidelines. LLM tier handles open-ended queries. Escalation triggers for sensitive medical topics.

E-Commerce Support

Product inquiries resolved by rules. Return policies grounded in knowledge base documents. Complex complaints handled by LLM with brand voice. VIP customers escalated to human agents.

SaaS Onboarding

Getting-started questions matched by rules. Feature documentation retrieved via RAG. Integration help handled by LLM with API docs context. Enterprise customers escalated to solutions engineers.

Comparison

Chat Agents vs. traditional chatbot builders

Dimension JieGou Chat Agents Traditional Chatbots
Matching Embedding-based cosine similarity Keyword / exact match
Knowledge RAG from uploaded documents None or manual FAQ
AI fallback Full LLM with conversation context None or basic templates
Escalation Configurable confidence threshold Manual rules only
Channels 5+ messaging platforms Usually 1-2
Governance RBAC, audit trails, sensitivity labels None

How We Compare

Chat Agents vs. the market

See how JieGou's 4-tier resolution cascade stacks up against other AI chat solutions.

Capability JieGou Zapier Chatbots Copilot Studio n8n AI Agents
Resolution Strategy Rules > RAG > LLM > Escalation (4 tiers) LLM-first (single strategy) LLM with Knowledge sources LLM with tool-level HITL
Token Efficiency 3 strategies before LLM (rules, RAG, escalation) Every message uses LLM tokens Every message uses LLM tokens Every message uses LLM tokens
Channels 12 channels (LINE, IG, WhatsApp, Messenger, Telegram, Slack, Teams, Discord, WeChat, email, web, SMS) Web embed + Slack Teams + web embed Web embed + custom
Knowledge Sources 13 enterprise adapters Zapier Tables SharePoint, OneDrive, web Custom via LangChain nodes
Governance 10-layer stack with agent identity and approval gates Basic admin roles Copilot Studio governance No governance layer
Multi-Model 9 providers per agent Zapier AI only Azure OpenAI only Multiple via LangChain

Build your first chat agent in minutes

Import rules from a CSV, connect a messaging channel, and start resolving customer messages automatically.