Usage

once brainz is live—doesn’t matter if you spun it up by hand or dropped it into docker—it’s all one stack. web ui, cli, and api share the same brain: one model state, one memory layer, one log stream. no silos, no bs.

---

1. web ui – talk to it, watch it think

fire it up: http://localhost (backend 8000 + frontend 3000 need to be running or containerized, obviously)

what you can mess with:

• throw prompts at the model, get live answers

• peek inside the memory layer (vector traces of past prompts)

• watch agents do their thing in real time (logs + metrics)

• trigger agents manually, force autotrain runs

• fine-tune directly in-browser without restarting anything

built for people who want to see what’s happening, not guess.

---

2. cli tools – for devs who hate clicking

lives here: backend/cli/

query.py – terminal inference

python cli/query.py --prompt "what is solana?"

output:

solana is a high-performance blockchain built for speed and scale...

train.py – fine-tune from the terminal

python cli/train.py \
  --prompt "describe lsts" \
  --completion "lsts are liquid staking tokens..."

why cli? because scripts > dashboards. easy to wrap in cronjobs, pipe data, or bulk-train from files.

---

3. api endpoints – automate everything

base url: http://localhost:8000 docs: http://localhost/api/docs (swagger ui for lazy days)

query the llm

POST /api/llm/query
{
  "prompt": "explain mev in ethereum"
}

response:

{ "response": "mev = maximal extractable value..." }

train live

POST /api/llm/train
{
  "texts": [
    "describe validator slashing in proof-of-stake systems."
  ]
}

stream logs

GET /api/system/logs?level=INFO&limit=100

returns every move the system makes, raw.

---

end-to-end flow – how it actually plays out

1. you push a prompt (cli, ui, or api – doesn’t matter)

2. agent checks the answer, flags it if trash

3. autotrain fires up, runs a quick fine-tune loop

4. prompt embedding gets stored in vector memory

5. next time a similar prompt hits → brainz recalls context instantly

6. you watch all this go down live in the ui logs

---

your workflow, your rules

monitor in the ui, script in the cli, integrate via api – all wired into the same runtime. agents retrain automatically, memory keeps evolving. you want to rewrite agent logic or hack the memory scoring? go ahead, it’s all yours.