Skip to content

QUICKSTART.md

Latest commit

 

History

History
48 lines (33 loc) · 2.32 KB

QUICKSTART.md

File metadata and controls

48 lines (33 loc) · 2.32 KB
Raw

Noetica Quickstart — zero to chatting

One command. From a clean machine to a running, local-first AI you can chat with.

bash scripts/install.sh

That's it. The installer is idempotent — re-run it any time, it only does what's missing.

What it does

  1. Preflight — checks your OS and git / node ≥ 20 / curl, printing the exact install hint for anything missing.
  2. Ollama — installs Ollama if you don't have it, starts the daemon, and pulls a sensible default model (qwen2.5:7b) plus embeddings (nomic-embed-text).
  3. Dependencies — npm install for the app and the agent-machine backend.
  4. Brain (optional) — offers to fetch a default brain from gs://noetica-brains if gcloud is installed and authed. If it isn't, it skips cleanly — verified-compute and reasoning work with no brain at all; only retrieval needs one.
  5. Launch — starts the headless agent-machine backend on :8080 and health-checks http://localhost:8080/api/status.
  6. Done — prints the URL and a ready-to-paste chat command.

Flags

bash scripts/install.sh --small      # smaller model (llama3.2:3b) for low-RAM boxes
bash scripts/install.sh --no-brain   # skip the optional brain fetch
bash scripts/install.sh --no-input   # never prompt (auto in CI / non-TTY)
bash scripts/install.sh --help       # full usage

NOETICA_MODEL=qwen2.5:14b bash scripts/install.sh   # pick any Ollama model
NOETICA_AM_PORT=9090       bash scripts/install.sh   # change the backend port

Chat

Once it's up, talk to the backend directly:

curl -N http://localhost:8080/api/chat \
  -H 'content-type: application/json' \
  -d '{"messages":[{"role":"user","content":"What is 17 * 23, and prove it?"}]}'

Responses stream back as Server-Sent Events. The Tauri desktop app (the GUI) talks to this same backend — run the installer once and the app picks it up.

The one thing nobody else does

Noetica is the only local AI that proves its computed answers. Every other local-AI and RAG system — Open WebUI, AnythingLLM, Jan, GPT4All, NotebookLM, Perplexity — shows you where text came from. None of them can show that an answer is correct. Grounding is not verification. That gap is structural, and it's ours alone.

See docs/COMPETITIVE.md for the honest head-to-head.