Nemoryn

English | Čeština | Deutsch

OpenAI-compatible memory server. Postgres runs in Docker. You bring your own chat model (Ollama or OpenAI).

Console: http://localhost:5022/
Open WebUI: http://localhost:5022/v1

What you need

• Docker Desktop (or a daemon) with docker compose
• Ollama (or an OpenAI-compatible API) with a chat model and an embedding model
• Port 5022 free. Compose maps Postgres to host port 5433 so it does not clash with local 5432

Start

git clone https://github.com/Mattes22/Nemoryn.git
cd Nemoryn
cp .env.example .env

In .env set at least:

POSTGRES_PASSWORD=your-password
MEMORY_AI_BASE_URL=http://host.docker.internal:11434
MEMORY_AI_CHAT_MODEL=gpt-oss:20b
MEMORY_AI_EMBEDDING_MODEL=nomic-embed-text

Ollama on another machine on the LAN: MEMORY_AI_BASE_URL=http://192.168.x.x:11434.

docker compose up --build -d

Open http://localhost:5022/. In Runtime, check that the database and the model are live. Inside Docker the database host is postgres, port 5432 (not 5433).

Stop: docker compose down. Volumes are removed only with -v.

Open WebUI

Admin → Settings → Connections → OpenAI:

Field	Value
API Base URL	http://127.0.0.1:5022/v1
API Key	anything if NEMORYN_API_KEY is empty; otherwise the same string

The model and the Ollama base URL are set in Nemoryn Runtime, not in Open WebUI.

Out of scope

• Open WebUI is not a plugin host. Candidates, conflicts, pin, and audit live in the console only.
• The console has no login. Do not expose 5022 to the public internet without a proxy or VPN.
• .env and memory-*.connection.json are not committed.

More on Docker: docs/docker.md. Open WebUI: docs/openwebui.md.