Workflow & AI — what brings processes and intelligence
The productive layer on top of the infrastructure: orchestrate workflows, operate models — both without client or personal data wandering off into third-party clouds.
Why this cluster
This is where the SMB's own intelligence enters: takes care of recurring processes (email triggers, , DATEV export, Slack bridges), is the ChatGPT-like interface on top of locally running models. The two are tightly coupled — can call via , can use workflows as tool calls. Together they form an stack you actually control.
Tools in this cluster
Ollama
Local LLM server with OpenAI API
Licence cost
Free
MIT licensed server for 100+ open-weight models (Llama, Mistral, Qwen, ): OpenAI-compatible , acceleration (NVIDIA/AMD/Apple), quantisation for RAM/VRAM efficiency. Example: how a German machine-building company with 80 staff replaced shadow IT (clandestine ChatGPT use) with a local server — IP protection instead of data leakage.
Open WebUI
Self-hosted chat interface for your own AI models
Licence cost
Free
Browser interface for local models (Llama, Mixtral, Qwen) via : multi-user with , , over your own documents, OpenAI-compatible . Example: how a German law firm with 8 staff runs § 203-compliant brief drafting, contract analysis and case-law research — without client data ever leaving the firm.
n8n
Workflow automation, self-hosted
Licence cost
Free
Fair-code tool for email triggers, , DATEV export, Slack bridges and CRM integrations. 400+ pre-built nodes, visual editor, runs on your own server. Example: how a tax firm with 60 clients automates its document intake, monthly P&L dispatch and deadline watchdog.
Camunda 7
BPMN workflow and DMN decision engine
Licence cost
Free
Apache-2.0 licensed engine + DMN decision engine in Java with user task inbox and cockpit. Example: how a German special machine builder with 60 staff turns engineering approval workflows from email threads into processes with user tasks, parallel paths and audit trail.
Stacks of multiple components
Tools are the building blocks — stacks are the finished backend platforms.
Stack
KI-Modell-Zoo
7 open-weight models compared
7 Modelle · Apache-2.0 + MIT + Custom Lic.
Llama 3.3, Mistral/Mixtral, Qwen 2.5, 3, DeepSeek R1, Phi 4, GPT-oss — overview with size, licence, VRAM footprint and recommendation. Plus cloud frontier comparison (ChatGPT, Claude, Gemini).
→ See the model zooStack
Supabase Stack
13 components as Backend-as-a-Service
13 Container · Apache-2.0 + MIT + MPL-2.0
+ PostgREST + GoTrue + Realtime + Storage + Studio + Kong + Supavisor + Edge Runtime + imgproxy + Logflare + Meta + Vector. Example: how an EdTech startup builds a GDPR-compliant platform for vocational schools — without Firebase.
→ See the stack showcaseOther clusters
Workflows need a platform and a surface
The tools in this cluster run on the infrastructure layer and talk to the user apps:
Ready for the next step?
Free intro call, no strings attached. In 30 minutes you'll know whether and how AI can help your business.