# Local Chat starter · v0.1

Run a small chat workspace on your own computer with Ollama and Open WebUI.
This free example comes from Sushii Kitchen's `combo.chat-local`. No Kitchen
account, Kitchen Credits, subscription or hosted model key is needed.

The archive includes configuration and instructions. Docker downloads the
upstream applications; you separately choose and download model weights.
The image identities are fixed in `compose.yaml` and `provenance.json`.

## Before you start

Use Docker with Compose v2 on a Mac or Linux computer. This starter runs on
CPU, including Apple Silicon through Docker's Linux VM. It does not enable
Metal or NVIDIA acceleration. The containers are limited to 4 GiB combined;
allow additional memory for Docker, your OS and other applications. Large
models need a different configuration. First downloads can take several minutes
and several gigabytes. We have not established a general setup-time or hardware
minimum.

This example exposes Open WebUI only at `127.0.0.1:18060`. Ollama has no
published host port. Keep the interface on your own computer; remote access
needs a separately configured authentication, TLS and network design.

## Start

1. Extract the archive and open a terminal in the `local-chat-v0.1` folder.
2. Create a private session secret, keeping any existing `.env`:

   ```sh
   (umask 077; set -C; printf 'LOCAL_CHAT_SECRET=%s\n' "$(openssl rand -hex 32)" > .env)
   ```

   If `.env` already exists, retain it. The command intentionally refuses to
   overwrite it. Never share this file. To use another loopback port, add
   `LOCAL_CHAT_PORT=18061` to the file.

3. Read the configuration, then start the two services:

   ```sh
   docker compose config --quiet
   docker compose up -d --wait --wait-timeout 360
   ```

4. Download a small model for the first connection check:

   ```sh
   docker compose exec ollama ollama pull smollm2:135m
   ```

   This is a small demonstration model, not a quality recommendation for your
   project. Model tags can change; record `docker compose exec ollama ollama list`
   and review the selected model's terms. Larger models can exceed this
   starter's memory limit.

5. Open **http://127.0.0.1:18060**, create the initial local administrator
   account, select the downloaded model, and send a short test message.
   Later public signups are disabled. This account belongs to your local
   Open WebUI installation, separate from your Sushii Kitchen account.

Success means the interface returns a model response and you can find that
conversation again after restarting. A healthy container alone is not that result.

## Return and keep your work

```sh
docker compose restart
docker compose ps
```

Your model files and local account/chat data live in separate named Docker
volumes. Sign in again and open your saved conversation. `docker compose down`
stops and removes this stack's containers and network while preserving volumes.
Restart with `docker compose up -d --wait --wait-timeout 360`.

Back up your data before upgrades or removal. **`docker compose down --volumes`
also deletes this project's local accounts, chat history and downloaded models.**
Use that command only when you deliberately want a clean removal. Do not use a
global Docker prune to remove this starter.

## Data and network choices

The configured chat connection is Open WebUI → Ollama in the same Compose
project. Hosted OpenAI-compatible connections and direct browser connections
are disabled; Open WebUI offline mode and update-check suppression are enabled.
Initial image/model downloads need internet access. These settings are not an
air-gap or a network-level egress guarantee. Adding integrations, tools, remote
models or web search changes the data path. Use non-sensitive sample text for
the first trial and inspect settings before using private material.

Tsugiita in the Sushii Kitchen portal is separate assistance. A locally running
Recipe does not imply that portal assistance runs locally. Review its configured
provider and the context it receives before sharing a project.

## Adapt the Recipe

- Inspect the components: https://www.sushikitchen.io/builder?from=combo.chat-local
- Save and review your project after sign-in:
  https://kitchen.sushii.io/dashboard/builder?from=combo.chat-local

Builder edits do not rewrite this downloaded example. Keep the example version,
your saved Recipe revision and any local changes identifiable. Proactive
maintenance and automatic updates are not included in this starter.

## Troubleshooting and terms

If startup times out, run `docker compose ps` and inspect bounded local logs:
`docker compose logs --tail=60`. Avoid sharing logs containing account or prompt
data. If a model is unavailable, confirm the download completed and that
Open WebUI's Ollama connection is `http://ollama:11434`. If a larger model fails,
try the small connection-check model before changing resource limits.

Sushii's example configuration is covered by the included Apache-2.0 license.
Upstream applications and model weights retain their own terms and notices:
[Ollama](https://github.com/ollama/ollama/blob/v0.32.9/LICENSE),
[Open WebUI](https://github.com/open-webui/open-webui/blob/v0.11.0/LICENSE),
and the [model listing](https://ollama.com/library/smollm2:135m).
Keep upstream branding and notices. The archive does not redistribute container
images or model weights and is not a managed-service or maintenance commitment.

The starter defaults to basic chat without native tool calling so the small
connection-check model works in the browser. Tool use requires a compatible
model and an explicit configuration change.