jeves-desk
Customer-service platform rebuilt from OpenAI's ChatKit sample where one Jev call decides intent, emotion and handling agent.
Overview
jeves-desk is a general customer-service platform refactored from OpenAI's ChatKit customer-support sample, with the airline-specific logic removed so that switching industries means editing configuration, not code. A ChatKit chat UI talks to a customer gateway that orchestrates without business rules. Jev acts as the brain: in one call it decides the customer's intent, emotion, confidence, which agent persona should handle the conversation (for example order service, product advisor or knowledge-base support) and which function is recommended.
A generative model then runs a function-calling loop over MCP data access, with code validating arguments with pydantic and stripping forged customer ids, and tool results fed back until a final reply is produced. RAG knowledge lookup and plugin tools are supported, a routing policy module applies confidence, emotion and confirmation thresholds, and business configuration lives in a single YAML file validated at startup. The ChatKit front-end assets are snapshotted locally so it works on networks where the CDN is blocked. There are 70 offline unit tests. Documentation is in Chinese, and it retains the original sample's licence and attribution.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What is the customer's intent? | — |
| Choice | Which agent persona should handle this conversation? | — |
| Choice | What emotion is the customer expressing? | — |
How it works
- Customer message arrives through ChatKit
- One Jev call decides intent, emotion, confidence, agent and suggested function
- Policy applies confidence and emotion thresholds
- LLM runs a validated MCP function-calling loop and replies
Strengths
- Industry change via YAML only
- Code validates tool arguments and strips forged ids
- 70 offline tests
Caveats
- Derived from a sample; production hardening unclear
- Chinese documentation
- Custom licence
At a glance
- Type
- MCP server
- Status
- beta
- Language
- Python
- Runtime
- FastAPI + React (ChatKit)
- Has code
- yes
- Has tests
- yes
- License
- Custom license
Use cases and integrations
How curated lists describe it
This repository implements a configurable customer-service platform combining ChatKit UI, Jev decision-making, generative chat, RAG knowledge lookup, plugin Tools, MCP data access, and YAML-driven Agent dispatch.
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev decide?
Intent, emotion, confidence, which agent handles the chat and the recommended function, in one call.
What does the LLM do?
Function calling over MCP data and writing the final reply.
How do I adapt it to a new business?
Edit the business YAML configuration rather than code.
Related projects
DeepSeek Harness plugin that classifies each user message's emotion, urgency and department with Jev and injects it as context.
PluginbetaTypeScriptZero-dependency Android app that judges the visible WeChat message's intent, emotion, urgency and reply posture with Jev.
AppbetaJavaAndroid WeChat assistant where Jev judges the chat and scores the relationship, and a chat model drafts three replies.
AppbetaKotlinWeChat plugin that classifies a long-pressed message's intent, emotion, urgency and suggested reply posture with Jev.
PluginbetaJavaLocal chat analyser that labels each WeChat message's emotion and intent with Jev and grades how your replies landed.
ProjectprototypeJavaScriptPython example classifying customer support messages into 12 intents with Jev, flagging clarification below 0.8 confidence.
ProjectprototypePython