<div align="center">
# Redis Iris Demos
**Reusable demo apps powered by Redis Iris**
Multi-domain AI agent demos for [Redis Iris](https://redis.io/iris/): Redis's unified context engine for AI agents.
Every domain runs Context Retriever, Agent Memory, LangCache, and Semantic Routing.
</div>
---
## Iris Components in Demo
| Solution | What it does |
|--------|-------------|
| **Semantic Routing** | Blocks off-topic queries before they reach the LLM |
| **LangCache** | Returns cached responses for semantically similar questions |
| **Agent Memory** | Enriches prompts with short-term session context + long-term user knowledge |
| **Context Retriever** | Give agents clean, schema-first paths through all business data |
## Domains
| Industry | App | Screenshot |
|----------|-----|------------|
| Food delivery | Redis Eats | <img src="docs/screenshots/Demo_Redis_Eats.png" width="300" /> |
| Electronics Retail | ElectroHub | <img src="docs/screenshots/Demo_ElectroHub.png" width="300" /> |
| Financial research | ShiftIQ | <img src="docs/screenshots/Demo_ShiftIQ.png" width="300" /> |
| Healthcare | RedHealthConnect | <img src="docs/screenshots/Demo_RedHealthConnect.png" width="300" /> |
| Retail banking | Radish Bank | <img src="docs/screenshots/Demo_Radish_Bank.png" width="300" /> |
| Telco | R-Mobile | <img src="docs/screenshots/Demo_R-Mobile.png" width="300" /> |
| Sports betting | Sports Desk | <img src="docs/screenshots/Demo_Sports_Desk.png" width="300" /> |
## Legacy
The `legacy/context-engine-demos/` directory contains an archived copy of the original `redis/context-engine-demos` repository — the first-generation demo that showcased Context Retriever only. The current repo supersedes it with Semantic Routing, LangCache, Agent Memory, and domain-specific branding.
## Getting Started
### Prerequisites
- Python 3.11+
- Node.js 18+
- [uv](https://docs.astral.sh/uv/) (Python package manager)
### Redis Cloud services needed
| Service | What you need |
|---------|--------------|
| Redis database | Host, port, password |
| [Context Surfaces](https://redis.io/docs/latest/develop/ai/context-engine/context-retriever/) | Admin key |
| [Agent Memory](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/) | Base URL, store ID, API key |
| [LangCache](https://redis.io/docs/latest/develop/ai/context-engine/langcache/) | Host, cache ID, API key |
| OpenAI | [API key](https://platform.openai.com/api-keys) |
### First time
```bash
make install # Install Python + JS dependencies
cp .env.example .env # Fill in credentials
make setup # Generate data, load Redis, seed memory + cache
make dev # Start at localhost:3040
```
### Coming back (same domain)
```bash
make dev # Data persists in Redis
```
If the agent has no tools (Redis was flushed, instance recycled), reload:
```bash
make reset # Reload data + re-seed
make dev
```
### Switching domains
```bash
# Update the DEMO_DOMAIN in .env
make setup
make dev
```
When switching to `radish-bank`, leave `MEMORY_SIMILARITY_THRESHOLD` unset or commented out
to use the Radish Bank Agent Memory default of `0.5`. Older local `.env` files may still
contain `MEMORY_SIMILARITY_THRESHOLD=0.7`; keep that only when you want to explicitly
override the Radish tuning.
## What `make setup` does
1. Sets `DEMO_DOMAIN` in `.env`
2. Generates Context Surface models from the domain schema
3. Generates synthetic data (JSONL files)
4. Flushes Redis database
5. Creates a new Context Surface (writes `CTX_SURFACE_ID` + `MCP_AGENT_KEY` to `.env`)
6. Loads data into Redis (creates search indexes)
7. Seeds Agent Memory (clears existing, seeds 2 long-term memories per domain)
8. Seeds LangCache (flushes cache, seeds 1 cached response per domain)
9. Semantic Routing initializes automatically when the server starts
`make reset` does steps 4-9 only (faster — reuses existing models and data files).
## Creating a New Domain
Adding a new vertical (e.g., telecom, travel, insurance). Every domain follows the same structure, so the process is repeatable.
### What you need to create
| File | What goes in it |
|------|----------------|
| `domains/<id>/schema.py` | Redis data entities (customers, orders, products, etc.) |
| `domains/<id>/domain.py` | Manifest: branding, theme, guardrail routes, seed memories, seed cache |
| `domains/<id>/prompt.py` | Agent system prompt with tool hints and workflows |
| `domains/<id>/data_generator.py` | Synthetic demo data (realistic names, amounts, scenarios) |
| `domains/<id>/assets/logo.svg` | Domain logo (renders at 28x28 in the topbar) |
| `domains/<id>/docs/demo_paths.md` | 2-4 scripted demo conversation paths |
| `frontend/public/backgrounds/<id>/left.svg` | Landing page left illustration (~590x817, <35KB) |
| `frontend/public/backgrounds/<id>/right.svg` | Landing page right illustration (~632x817, <35KB) |
Each domain implements the `DomainPack` protocol defined in `backend/app/core/domain_contract.py`.
### Using AI to create a domain
The fastest way to create a new vertical is with an AI coding assistant. Detailed authoring skills are available for both:
- **Claude Code**: `.claude/skills/domain-pack-authoring.md`
- **Codex**: `.codex/skills/domain-pack-authoring/SKILL.md`
Just ask: *"Create a new domain for [your vertical]"* — the skill guides the agent through the full process: entity design, branding, realistic guardrail routes, seed data, prompts, and demo scenarios.
## Project Structure
```
backend/
app/
main.py # FastAPI app, SSE streaming, 7-phase pipeline
langgraph_agent.py # LangGraph ReAct agent
memory_service.py # Redis Agent Memory client
context_surface_service.py # MCP tool wrapper
guardrail_service.py # Semantic routing guardrail
langcache_service.py # Semantic cache client
settings.py # Pydantic BaseSettings
core/
domain_contract.py # DomainPack protocol + configs
domain_schema.py # EntitySpec definition
domains/
reddash/ # Food delivery
electrohub/ # Electronics retail
finance-researcher/ # Financial research
healthcare/ # Patient portal
radish-bank/ # Retail banking
rmobile/ # Wireless telecom
sports-betting/ # Sports betting
frontend/
src/
App.tsx # React chat UI, SSE handler
components/ # ConversationView, ActivityPanel, EmptyState
scripts/ # Setup, data loading, seeding, validation
.codex/skills/
domain-pack-authoring/ # AI skill for creating new domains
```
## Make Targets
| Target | What it does |
|--------|-------------|
| `make domains` | List available domains |
| `make setup [DOMAIN=X]` | Full setup: models, data, Redis, seed |
| `make reset` | Reload current domain into Redis + re-seed |
| `make dev` | Start backend (:8040) + frontend (:3040) |
| `make install` | Install Python + JS dependencies |
| `make create-domain DOMAIN=X` | Scaffold a new domain |
| `make seed-memories` | Re-seed long-term memories for current domain |
| `make seed-langcache` | Re-seed LangCache entries for current domain |
| `make flush-redis` | Wipe Redis database |
All targets read `DEMO_DOMAIN` from `.env` — no need to pass `DOMAIN=` unless switching.
---
## License
MIT