Chatbot Tutorial

SymbiaChat is the built-in conversational agent that ships with SymbolicAI. You don’t have to derive your own ChatBot subclass anymore—everything is already wired together for you.

---

Quick start

1 · From the terminal

symchat

2 · From Python

from symai.chat import SymbiaChat

chat = SymbiaChat() # optional: verbose=True for debug prints
chat()

Both entry points start an interactive REPL. Type your messages and Symbia will reply in real time.

---

What can Symbia do?

Symbia continuously classifies your input and decides whether she can answer directly, needs a tool, or should ask for clarification. Below is the current toolbox with the classification tags it uses internally (you will occasionally see them in debug mode):

Tag	Engine	Typical prompt examples
[WORLD-KNOWLEDGE]	Internal LLM	“Who wrote Dune?”
[SYMBOLIC]	Symbolic math solver	“Integrate x² from 0 to 3.”
[SEARCH]	Web search (RAG)	“Give me the TL;DR of arXiv:2507.16075.”
[SCRAPER]	Website scraper	“Scrape all contributor names from .”
[SPEECH-TO-TEXT]	Audio transcription	“What does this file say? /path/to/audio.mp3”
[TEXT-TO-IMAGE]	Image generation	“Draw a friendly space cat.”
[FILE]	File reader	“Summarise /tmp/report.pdf.”
[RECALL]	Memory (short + long term)	“Remember that my dog’s name is Noodle.”
[HELP]	Capability list	“What can you do?”
[DK]	“Don’t know” ↔ ask user	Ambiguous or malformed input
[EXIT]	Quit session	“Goodbye”, “quit”, …

---

Memory model

1. Short-term memory – a fixed-size sliding window that keeps the last n conversation turns.

2. Long-term memory – a vector index stored on disk (localdb/<index>.pkl) for facts Symbia deems worth remembering. Ask “Do you remember …?” to trigger a [RECALL].

Both memories are consulted automatically; you do not need to manage them.

---

Under the hood (tl;dr)

ChatBot                # abstract base class
└── SymbiaChat         # concrete subclass you interact with
    ├─ In-context classification → picks capability tag
    ├─ SlidingWindowListMemory  → short-term memory
    ├─ Vector index interface   → long-term memory
    └─ self.interfaces[...]     → tooling shown above

If your use-case demands different behaviour, subclass ChatBot and override forward, but for most applications SymbiaChat is already sufficient.

---

Next steps

• Start a chat, explore the capabilities, and watch the classification tags in verbose mode for insight into Symbia’s decision making.

Happy chatting!