Free starter · Local Chat v0.1

Your first local chat project.

Ollama runs a model. Open WebUI gives you a chat screen. Download the starter and run both on your own machine, with no Sushii account, subscription, conversation credits or provider key required.

Download Local Chat Explore this Recipe

CPU example for Docker Compose v2 on Mac or Linux. The two containers have a combined 4 GiB memory limit; Docker and your operating system need additional memory. Expect several gigabytes of downloads and several minutes of setup. Performance depends on your machine and model.

Install and try a conversation

Read the included README first. In a terminal, open your download folder and run these commands. The secret command creates a private .env file and refuses to overwrite an existing one.

unzip local-chat-v0.1.zip
cd local-chat-v0.1
(umask 077; set -C; printf 'LOCAL_CHAT_SECRET=%s\n' "$(openssl rand -hex 32)" > .env)
docker compose config --quiet
docker compose up -d --wait --wait-timeout 360
docker compose exec ollama ollama pull smollm2:135m

Open 127.0.0.1:18060, create your local administrator account, select smollm2:135m and send a short sample message. This tiny model checks that the connection works; choose and assess a suitable model before relying on its answers. The starter uses basic chat without native tool calling.

Run docker compose restart, sign in again and reopen the chat. Models and chats live in named Docker volumes. Use docker compose down to stop the stack while retaining them. The README explains backups, cleanup and troubleshooting.

Full setup guide File checksums Pinned images and provenance

Know where your project runs

This profile binds the chat screen to this machine only and uses local Ollama inference. Installation and model downloads need internet access. Its offline settings are not a network firewall. Start with sample text and assess your own privacy requirements before using private data. GPU acceleration and remote access are outside this starter.

The archive contains configuration and instructions, not container images or model weights. Each upstream project and model has its own terms, linked in the README. Keep Open WebUI branding intact.

Turn it into your own Recipe

The public Builder lets you inspect components and their requirements. Tsugiita roughly translates into the concept that we have of a “Sous Chef.” For a saved Recipe and Tsugiita assistance, continue in the signed-in Kitchen workspace. You can also start there with a blank project idea. Changes in the Builder do not rewrite this downloaded archive. Builder resource estimates follow catalog defaults, including a GPU profile for Ollama. This download uses a separate CPU profile; follow its README for the configuration and limits.

Describe your goal, machine, privacy needs, budget and maintenance preferences in the project brief.
Ask Tsugiita, inspect its proposal and explicitly apply the components you want. Save the Recipe and reopen its saved page.
On a later visit, change one requirement, save the brief as a new revision and request a review of that saved Recipe.

Tsugiita sends your question and relevant project context to the configured AI provider. Running this chat starter locally does not make Kitchen assistance local. Use non-sensitive project examples. Save important requirements in the brief; the preview does not retain a conversation history. Saved-Recipe recommendations are advisory; applying architecture changes and scheduled Pro reviews are not provided by this starter.

Continue in Kitchen