Workshop materials for VibeKode Munich 2026, session Connecting AI.
Trainers: Piotr Zalewa & Michał Komorowski (DeepFellow core team)
A working DeepFellow proxy that sits between your application and one or more LLM backends. By the end of the session you will have wired up model routing, access control, observability, and a production-ready API endpoint, with running code you can carry back to your own stack.
No prior DeepFellow experience required. Familiarity with the command line is enough.
DeepFellow is a modular framework made of components. Each component is independent. Use one. Use a few. Use them all. Combine them the way your architecture demands. Scale up, scale down, but never lose control over your data and costs.
Production-ready infrastructure for self-hosted AI deployments. Full data sovereignty, built-in compliance, OpenAI-compatible API.
Because the trainers built DeepFellow, you get answers straight from the source.
-
bore (see below)
# macOS
brew install bore-cli
# Cargo (all platforms)
cargo install bore-cli
# Arch Linux
yay -S bore
# Gentoo
sudo emerge net-proxy/bore
# Prebuilt binaries (Linux, macOS)
# https://github.com/ekzhang/bore/releases# CLI install
curl https://deepfellow.ai/install.sh | bash
# Run the CLI
deepfellow
# DFServer install
deepfellow server install
# Start the DFServer
deepfellow server start
deepfellow server create-admin
# Local DFInfra install
deepfellow infra install
# Expose DFInfra via bore
# In separate terminal
bore local 8086 --to bore.pub
# Update the INFRA_URL
deepfellow infra env set INFRA_URL http://bore.pub:4321
deepfellow infra info
# Connect Infras
deepfellow infra connect wss://<YOUR TEAM INFRA>.workshop-demo.deepfellow.com <DF_MESH_KEY>
# df_infra_tap plugin
cp -r df_infra_tap ~/.deepfellow/server/plugins/
deepfellow server restart
# in a separate terminal
deepfellow server logs -f | grep "INFRA TAP"
# PII anonymization
deepfellow server env set PLUGINS_SETUP '{"df_anonymize_models": ["gpt-5-mini"]}'
# Audit — vector stores (Block 2)
deepfellow otel logs -f
# MCP Patient database (Block 3)
# In patient-database-mcp/ directory
docker compose up -d --build
# Audit — MCP (Block 3)
deepfellow otel logs -f --tail 1000{
"input": "Search files for psoriasis drug that achieved complete skin clearance in clinical studies",
"model": "gemma4:e4b",
"tools": [
{
"type": "mcp",
"server_label": "workshop",
"server_url": "df://local/toolboxes/YOUR_TOOLBOX_ID/mcp",
"require_approval": "never"
}
]
}# first question
Search files for psoriasis drug that achieved complete skin clearance in clinical studies
# second question
Which patients in our database are at risk for psoriasis?