diff --git a/.dockerignore b/.dockerignore
new file mode 100644
index 0000000..f33487b
--- /dev/null
+++ b/.dockerignore
@@ -0,0 +1,17 @@
+.git
+.gitignore
+.github
+.venv
+.glama-venv
+.pytest_cache
+__pycache__
+**/__pycache__
+*.py[cod]
+.fastembed-cache
+.planning
+benchmarks
+tests
+assets/*.mp4
+docs
+examples
+*.log
diff --git a/Dockerfile b/Dockerfile
new file mode 100644
index 0000000..65f47ed
--- /dev/null
+++ b/Dockerfile
@@ -0,0 +1,35 @@
+# Container image for the stdio MCP server.
+#
+# Glama builds this to introspect the 29 tools and to offer a deployable image;
+# `docker build -t turbo-quant-memory . && docker run -i --rm turbo-quant-memory`
+# is the same thing by hand. The server speaks MCP over stdin/stdout, so the
+# container is meant to be driven by an MCP client, not published on a port.
+FROM python:3.13-slim
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    PIP_NO_CACHE_DIR=1 \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    TQMEMORY_HOME=/data \
+    TQMEMORY_DAEMON_DISABLE=1 \
+    TQMEMORY_MIGRATE_ON_STARTUP=1
+
+# onnxruntime (through fastembed) and pyarrow (through lancedb) load libgomp at
+# import time; the slim image does not carry it.
+RUN apt-get update \
+ && apt-get install -y --no-install-recommends libgomp1 \
+ && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+COPY . .
+RUN pip install .
+
+RUN useradd --create-home --uid 1000 tqmemory \
+ && mkdir -p /data \
+ && chown -R tqmemory:tqmemory /data
+USER tqmemory
+
+# Memory lives in /data; mount a volume to keep it across runs.
+VOLUME ["/data"]
+
+CMD ["turbo-memory-mcp", "serve"]