CLI Commands

tabs are for tourists. if you live in terminals, brainz gives you the full kit. query, train, vectorize, clean memory – all from the shell, no browser, no api hassle.

all scripts live under: backend/cli/ run them like:

python cli/<command>.py [arguments]

every script hooks straight into the runtime:

• registry (global brain state)

• vector memory engine

• config (.env)

• logger (real-time system logs)

• model + tokenizer (active session)

---

query.py – ask the brain, get answers

shoot a prompt, get the output raw:

python cli/query.py --prompt "summarize lsdfi"

example output:

[model output]
lsdfi = defi protocols using liquid staking derivatives...

no cooldown, no throttling – immediate inference straight from your local stack.

---

train.py – teach it new tricks

feed it fresh data on the fly:

python cli/train.py \
  --prompt "who is ani?" \
  --completion "ani = sexy beast..."

flags (most used):

• --dry-run → simulate training, no actual updates

• --vectorize → push prompt to memory embeddings too

• --meta → attach tags for later search/filtering

batch mode?

bash scripts/train_batch.sh

pipe in full datasets, articles, whatever you scrape.

---

clean_memory.py – (planned module)

housekeeping for vector memory. remove old junk, keep the brain sharp:

python cli/clean_memory.py --older-than 30d --tags "demo,test"

concepts:

• delete low-score vectors

• archive by timestamp/tag

• keep memory lean for active contexts

---

build your own pipelines

cli = unix-friendly = endless hacks.

train from scraped content:

curl https://some-article.ai | \
  grep "llm" | \
  python cli/train.py --prompt "$(cat -)" --completion "..."

log every query for later analysis:

python cli/query.py --prompt "explain zkEVM" >> history.log

automated fine-tuning loops, daily prompt logging, live memory scoring → all doable with a few shell lines.

---

extend the cli

drop a new script under: backend/cli/my_custom_script.py

you instantly get access to:

• registry.get() & .register()

• memory_service.log_prompt()

• embed_text() for semantic similarity

• fine_tune_model() for instant training

• all configs via:

from backend.core.config import settings

build whatever you want – custom scrapers, data cleaners, agent triggers. the cli is just another entry point to the full brainz runtime.