From bdb403812b68d61206c316e023c3701129e12452 Mon Sep 17 00:00:00 2001
From: fegger <flo.egger@icloud.com>
Date: Tue, 19 May 2026 23:33:07 +0200
Subject: [PATCH] Update README.md

---
 README.md | 57 +++++++++++++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 53 insertions(+), 4 deletions(-)

diff --git a/README.md b/README.md
index 0d2e294..8e852c7 100644
--- a/README.md
+++ b/README.md
@@ -59,9 +59,9 @@ Edit `.env.local`:
 DATABASE_URL=postgresql://user:password@localhost:5432/loam
 JWT_SECRET=change-this-to-a-long-random-string-in-production
 
-# Optional — defaults shown
+# Optional — recommended local model shown
 OLLAMA_URL=http://localhost:11434
-OLLAMA_MODEL=llama3.1:8b
+OLLAMA_MODEL=qwen3:32b
 ```
 
 ### 3. Set up the database
@@ -77,9 +77,11 @@ npm run db:seed
 ### 4. Pull an Ollama model (optional)
 
 ```bash
-ollama pull llama3.1:8b
+ollama pull qwen3:32b
 ```
 
+`qwen3:32b` is the recommended local model for the recommendation endpoint on a 32GB VRAM GPU such as the AMD Radeon AI PRO R9700. Use `qwen3:14b` if you want faster responses or need a smaller memory footprint.
+
 ### 5. Run the dev server
 
 ```bash
@@ -90,6 +92,53 @@ Open [http://localhost:3000](http://localhost:3000).
 
 ---
 
+## Docker
+
+Build and run the web app plus PostgreSQL:
+
+```bash
+docker compose up --build
+```
+
+The app is exposed at [http://localhost:3333](http://localhost:3333). The Compose stack provides `DATABASE_URL` and `JWT_SECRET` for the app container.
+
+The Docker image installs dependencies with `npm ci --ignore-scripts`; Prisma generation runs later in the builder stage after `prisma/schema.prisma` has been copied into the image. The base image also installs OpenSSL so Prisma's native query engine works on Alpine.
+
+### Ollama in a Separate Container
+
+The app expects Ollama at `OLLAMA_URL`. If Ollama runs in another container, expose port `11434` and point the app at that service or host address.
+
+For an AMD GPU with ROCm, a typical Ollama container launch is:
+
+```bash
+docker run --rm -it \
+  --device=/dev/kfd \
+  --device=/dev/dri \
+  --group-add video \
+  --group-add render \
+  -v ollama:/root/.ollama \
+  -p 11434:11434 \
+  ollama/ollama:rocm
+```
+
+Then pull the recommended model:
+
+```bash
+ollama pull qwen3:32b
+```
+
+If the app runs in Docker Compose and Ollama is bound to the host on Linux, set:
+
+```yaml
+environment:
+  OLLAMA_URL: http://host.docker.internal:11434
+  OLLAMA_MODEL: qwen3:32b
+extra_hosts:
+  - "host.docker.internal:host-gateway"
+```
+
+---
+
 ## Project structure
 
 ```
@@ -225,4 +274,4 @@ Android commands are run from `android/`:
 | `DATABASE_URL` | Yes | — | PostgreSQL connection string |
 | `JWT_SECRET` | Yes | — | Secret for signing session JWTs |
 | `OLLAMA_URL` | No | `http://localhost:11434` | Ollama server base URL |
-| `OLLAMA_MODEL` | No | `llama3.1:8b` | Model name for recommendations |
+| `OLLAMA_MODEL` | No | `llama3.1:8b` | Model name for recommendations. `qwen3:32b` is recommended for a 32GB VRAM local GPU. |