kstudiochat 1.0.0

K-CHAT — Universal Chatbot Engine. Anti-hallucination by construction.

K-CHAT — Universal Chatbot Engine

Zero-hallucination by construction.
Drop a folder of documents. Get a chatbot. No ML expertise needed.

pip install kchat
kchat init ./helpdesk --template government
kchat chat --data ./helpdesk

Three commands. Working bot. No API key required.

---

Quick Start

# Install
pip install kchat

# Create a bot from a pre-built template
kchat init ./my-bot --template restaurant

# Chat with it (uses SimulatedLLM — works offline, no API key)
kchat chat --data ./my-bot

# Or use OpenAI/DeepSeek
pip install kchat[openai]
# Edit my-bot/config.json → change llm.provider to "openai" or "deepseek"

Available templates

kchat templates

Template	Description
government	FAQ bot for passports, permits, taxes
restaurant	Menu info, hours, reservations
healthcare	General health information and clinic FAQ

---

How It Works

K-CHAT uses an Industry Pack — a folder of data files. No code changes needed to onboard a new domain.

my-industry-pack/
├── config.json       # Bot identity, LLM choice, retrieval settings
├── intents.json      # Intent taxonomy (optional)
├── refusal.json      # Fallback responses + escalation contact
├── knowledge/        # Documents the bot answers from
│   ├── overview.md
│   └── faq.md
├── context/          # Style guides, policies (optional)
├── SOUL.md           # Personality definition (optional)
├── skills/           # Custom Python workflows (optional)
└── tools/            # Custom API tools (optional)

The pipeline:

User Input → Sanitize → Ethics Check → Intent Classify → Emotion Detect
→ Cultural Adapt → Retrieve Knowledge → LLM (grounded prompt)
→ Anti-Hallucination Verify → Tone Apply → Response

Anti-Hallucination

K-CHAT never asks the LLM to recall facts. It:

1. Injects only retrieved context into the prompt
2. Verifies every response against the source documents
3. Falls back to extractive answers if verification fails

Three verification strategies: entity consistency, citation coverage, text overlap.

---

CLI Reference

Command	Description
kchat init <path>	Create a new industry pack
kchat init <path> --template <name>	Create from pre-built template
kchat templates	List available templates
kchat chat --data <path>	Interactive chat in terminal
kchat serve --data <path>	REST API server (stdlib HTTP)
kchat validate --data <path>	Validate pack structure
kchat curate --data <path>	Audit knowledge quality
kchat info --data <path>	Show resolved config
kchat audit --data <path>	Full self-audit

Chat REPL

(My Bot) You > /help
  /quit   - Exit
  /clear  - Clear screen
  /multiline - Enter multiline mode (end line with """)
  /help   - Show this help

(My Bot) You > What are your hours?
Bot: We are open Monday to Saturday: Lunch 11-2:30, Dinner 5-10.
  sources: faq.md | confidence: High (0.93) | intent: info_lookup

REST API

kchat serve --data ./my-bot --port 8000
curl -X POST http://localhost:8000/chat \
  -H 'Content-Type: application/json' \
  -d '{"message": "What are your hours?"}'

Endpoints: GET /health, POST /chat, WS /ws, GET /info

Auth: set KCHAT_API_KEY env var. Rate limit: set KCHAT_RATE_LIMIT (req/min).

---

Extensions

Enable via config.json:

{
  "extensions": {
    "live": {"enabled": true},
    "conversational_memory": {"enabled": true},
    "multimodal": {"enabled": true},
    "predictive": {"enabled": true}
  }
}

Extension	What it does
live	Injects real-time data into context (prices, queues, sensors)
conversational_memory	Tracks user preferences, mood, relationship stage
multiagent	Routes queries to specialist agents
multimodal	Enhances emotion detection (text + optional voice/visual)
adaptive	Tracks strategy effectiveness, recommends adjustments
predictive	Assesses risk based on mood patterns and conversation signals
rlhf	Collects feedback and computes reward signals
voice	STT → Bot.chat() → TTS pipeline

Import directly: from engine.extensions.voice import VoiceManager

---

Configuration

Minimal config.json:

{
  "name": "My Bot",
  "persona": "Helpful, accurate, and concise.",
  "llm": {"provider": "simulated"}
}

Available LLM providers: simulated (default, offline), openai, deepseek.

Full config reference: kchat info --data ./my-bot

---

Deployment

Railway

# Deploy from GitHub
# Set KCHAT_DATA=./my-bot
# Set PORT=8000

VPS / Docker

# Install, create pack, run server
pip install kchat
kchat init ./my-bot --template government
kchat serve --data ./my-bot --host 0.0.0.0 --port 8000

Vercel (serverless)

See clients/nail-art/ for a working Vercel deployment example.

---

Why K-CHAT?

Instead of...	K-CHAT gives you...
LangChain (framework)	A turnkey engine with a data contract
Building from scratch	Anti-hallucination out of the box
Paying per chatbot	One engine, infinite domains
Complex ML pipelines	Deterministic, auditable verification

---

License

MIT