# One Sentence Summary
The system combines **LLM reasoning, guardrails, tools, and a vector database** into a coordinated workflow managed by the **OpenAI Agents SDK Runner**.
🥐 Only ask questions about food
AI Agents in a Nutshell — Architecture
User
|
v
Chainlit
(UI + Authentication + Chat Interface)
|
v
OpenAI Agents SDK
Runner
(Orchestration Layer)
|
+------------------------+
| |
v v
Input Guardrail Agent
(topic filter) (LLM reasoning)
| |
| v
| Tools
| (calorie_lookup_tool)
| |
+------------------------+
|
v
ChromaDB
(Vector Database)
nutrition_db collection
|
v
Post-Processing Hook
(Croissant Upsell)
|
v
Final Response### Chainlit
Provides the **chat interface and authentication**.
Handles:
* message streaming
* UI steps for tool calls
* user login
---
### OpenAI Agents SDK
The **Runner** orchestrates everything:
* calls the agent
* enforces guardrails
* executes tools
* streams results back
---
### Agent
The AI reasoning layer.
It decides:
```
answer directly
OR
call a tool
```
---
### Guardrails
Input guardrails ensure the system only answers **nutrition-related questions**.
Example blocked question:
```
How many roads lead to Rome?
```
---
### Tools
Custom functions the agent can call.
Example:
```
calorie_lookup_tool()
```
Which queries the vector database.
---
### ChromaDB
Stores food data as **vector embeddings**.
Example:
```
Peanut Butter → 384-dimensional vector
```
Semantic search finds foods based on meaning rather than keywords.
---
### Post-Processing Hook
After the model finishes, a custom function modifies the response.
Example feature:
```
🥐 Croissant upsell for French cuisine
```
def croissant_upsell(text: str) -> str:
french_triggers = [
"france",
"french",
"paris",
"niçoise",
"provence",
"dijon",
"bordeaux",
"baguette",
]
if any(word in text.lower() for word in french_triggers):
text += "\n\n🥐 For only 120 calories more, may we interest you in a croissant with that?"
return text---
https://ai-agents-in-a-nutshell.onrender.com/
---
# Why This Architecture Matters
It demonstrates how modern AI applications are built:
```
LLM
+ tools
+ retrieval
+ orchestration
+ deployment
```
The model provides reasoning, but the **framework coordinates the system**.
---
AI Agents in a Nutshell — Architecture
+----------------------+
| User |
| Browser / Tester |
+----------+-----------+
|
v
+----------------------+
| Chainlit |
| UI + Auth + Chat |
+----------+-----------+
|
v
+----------------------+
| OpenAI Agents SDK |
| Runner |
| Orchestration |
+----------+-----------+
|
+-----+-----+
| |
v v
+---------+ +------------------+
| Input | | Agent |
| Guardrail| | LLM Reasoning |
+----+----+ +---------+--------+
| |
| v
| +------------+
| | Tools |
| | calorie |
| | lookup |
| +------+-----+
| |
+---------------------+
|
v
+------------------+
| ChromaDB |
| nutrition_db |
| vector search |
+--------+---------+
|
v
+------------------+
| Post-Processing |
| Croissant Upsell |
+--------+---------+
|
v
+------------------+
| Final Response |
+------------------+
🥐 Only answers question about food. Some questions about French food will try to upsell you a croissant. That is the bonus "Output Guardrail".