MIND - Modular Inference & Node Database

Version: Milestone-1

Setup

This will create and setup a SQLite database if mind.db doesn't exist under backend/.

llama.cpp is included as a third-party git submodule under third_party.

A script for grabbing a Qwen 3 1.7B LLM model from HuggingFace is at backend/get-qwen3-1.7b.sh.

The default location for the Go layer of the backend to access the LLM service is localhost:8081.

Please consider adding the -n <num-of-tokens-to-predict> option to avoid infinite generation on smaller models

Run go run ./backend/main.go, which will start a backend instance at localhost:8080.

Note that if mind.db doesn't exist in the working directory, it will be created and be initialized by the code in backend/internal/db/migrations.

All db-related operations (e.g. queries) are under backend/internal/db.

To run tests, we provide a simple CLI-based frontend client at backend/cli.py.

This will include all of the features supported or under development.

A short video demo of basic functionality on the CLI can be found under milestone-1/demo.mp4.

The specific APIs used are given in backend/design.md.

Action	Command / Description
Start conversation	`\new <title> <owner_id>`
Commit (prompt/answer or node)	Free text → `/completion` or `\node <user\|assistant> <content> [parent_id]`
Fork branch	`\branch <name> [head_node_id]`
Delete branch or node	`\delbranch <name>` or `\delnode <node_id>`
Extract subset as new tree	`\extract <title> <node_id ...>`
Merge two branches	`\merge <left> <right> <left-first\|right-first>`
Detach branch to new conversation	`\detach <branch>`
Append / copy tree to another	`\append-tree <src_conv> <src_branch> [dst_conv [dst_branch]]`
List conversations	`\listconvs <owner_id>`
List branches	`\listbranches <conversation_id>`
Linearize current branch	`\linearize`
Debug llama-server directly	`\llama <prompt>`
Switch conversation / branch	`\useconv <id>` and `\usebranch <name>`
Show current user	`\who`

For a more detailed experience, it's strongly recommended to run the backend locally and trying out the CLI.