talkatoo 0.1.0

Talkatoo is an offline CLI conversation engine with custom NLP, retrieval, and local memory.

Talkatoo

Talkatoo is a local CLI conversation engine built around custom NLP, session memory, document retrieval, and deterministic tools. It does not use any external API, HuggingFace, Torch, or chatterbot.

Stack

• Python 3.11 for the Talkatoo engine, memory, retrieval, planning, and CLI control flow
• JavaScript (Node) for terminal rendering and CLI presentation
• Bash for the Talkatoo launcher that clears the screen before startup

Features

• SQLite-backed session history and explicit fact memory
• Local document indexing from text, markdown, source files, and config files
• Custom NLP for topic extraction, relation detection, feedback repair, and context carry-over
• Dynamic response synthesis from parsed user input plus grounded evidence instead of fixed section templates
• Intent routing for explanation, planning, debugging, comparison, recall, and math
• Safe arithmetic tool with a small allow-list of math functions
• Slash-command driven CLI with session export and live knowledge loading
• Streamed assistant output through the Node terminal renderer
• Seed playbook content so the bot has grounded engineering guidance out of the box

Run

From the project root:

./run_talkatoo.sh

Or directly:

python3 -m talkatoo.app

Commands

• /help
• /clear
• /mode analyst|builder|tutor
• /session new [title]
• /sessions
• /facts [query]
• /load <path>
• /export [path]
• /quit

Notes

• run_talkatoo.sh clears the terminal before launching the CLI.
• /load <path> can index a local file or directory at runtime.
• Exported transcripts are written as markdown into data/ by default.
• Talkatoo is only as knowledgeable as the files you load plus the built-in playbook and stored session memory.