5-chatbot_agentic_v3.md•4.08 kB
# 🧠 Agentic Chatbot v3 — With Long-Term Semantic Memory
This third version builds on the agentic design of v2, but adds a powerful new layer:
**vector database integration** for semantic memory.
With this, the chatbot becomes more context-aware, capable of retrieving relevant past conversations even if the wording is different.
It simulates long-term memory, something that wasn't possible in the earlier versions.
---
## 🔍 Core Functionalities
### 1. 🧱 Initialization
As before, the chatbot loads all the foundational components:
- OpenAI client
- Configurations
- Session ID
- SQL-based managers
But now it adds:
- **`VectorDBManager`** — connects to a vector database (e.g., Chroma) for storing and retrieving embeddings of past conversations.
It also updates the list of tools (functions the agent can call) to include:
- `add_user_info_to_database`
- `search_vector_db` (replacing `search_chat_history` from v2)
---
### 2. ⚙️ Function Execution
The chatbot supports OpenAI function calling.
Now, when the model asks to search memory, it can do so **semantically** using vector similarity rather than keyword matching.
This leads to more relevant, nuanced, and context-aware behavior.
---
### 3. 🔁 Conversational Loop
Same looping architecture as v2:
- The chatbot keeps prompting the model until a final response is ready
- It handles tool calls, updates chat history, and builds a dynamic system prompt using:
- User info
- Chat summary
- Chat history
- Function call results
- And now: vector memory!
The system prompt uses a new builder:
`prepare_system_prompt_for_agentic_chatbot_v2` — designed to support semantic memory and vector search results.
---
### 4. 🧠 Vector Memory Update
After every conversation:
```python
msg_pair = f"user: {user_message}, assistant: {assistant_response}"
self.vector_db_manager.update_vector_db(msg_pair)
```
This message pair is embedded and stored in the vector database.
That means the next time the user asks a related question, the chatbot can recall this past exchange even if it's phrased differently.
---
## 🔄 Key Differences from the Previous Versions
| Feature | **V1: Basic Chatbot** | **V2: Agentic Chatbot** | **V3: Vector-Aware Agentic Chatbot (This)** |
|--------|-----------------------|--------------------------|--------------------------------------------|
| **Chat Flow** | Simple Q&A | Agent loop with function calls | Agent loop with memory + vector updates |
| **Function Calling** | ❌ None | ✅ Basic agent functions | ✅ Includes vector DB search |
| **Tools** | None | `add_user_info_to_database`, `search_chat_history` | `add_user_info_to_database`, `search_vector_db` |
| **Search Capability** | ❌ None | ✅ Keyword/phrase-based search on SQL database (not semantic) | ✅ Semantic search using vector embeddings (stored in a vector database) |
| **Prompt Strategy** | Simple prompt | Prompt with function call context | Prompt with function call context
| **Fallback Strategy** | ❌ None | ✅ Includes a fallback mechanism| ✅ Includes a fallback mechanism|
| **Memory System** | Chat history + summary | Chat history + summary + **SQL db search result** | Chat hirsoty + summary + **long-term memory (vector DB)** |
| **System Prompt** | `prepare_system_prompt` | `prepare_system_prompt_for_agentic_chatbot_v1` | `...v2` with vector support context |
| **Knowledge Retention** | Temporary, short-term | Contextually limited | **Semantically indexed for long-term reasoning** |
---
## 🧠 Final Thoughts
This chatbot represents a **hybrid agent** — blending dynamic tool use with smart, long-term memory.
It:
- Understands conversations across sessions
- Grows more intelligent over time
- Uses semantic reasoning to bring relevant context into its responses
This is the blueprint for building **truly intelligent assistants** — not just reactive models, but adaptive agents.
---
## 🧾 Architecture Overview
Here's the architecture of Agentic Chatbot v3:
