CoLab IA — How Three People Run a Community with One Vault
AI is better explored together.
My role
- Co-founder — strategy, content, community direction
- AI architect — designed the workspace, workflows, and publishing pipeline
- Organiser — events, communication, member engagement
What is CoLab IA?
CoLab IA is a collaborative community for AI experimentation, based in Lausanne, Switzerland. It brings together enterprises, institutions, and citizens to demystify artificial intelligence — through practice, not theory.
What happens at a CoLab IA event
- 30 min — demystification with concrete examples
- 60 min — hands-on workshop in small groups
- 30 min — sharing results and community building
Every participant leaves with:
- A clear understanding of what AI can (and cannot) do
- Tools they can reuse immediately
- An informed perspective on ethical implications
- Access to a local community of curious practitioners
Founded in 2025 by three people with complementary backgrounds: project management, training, and change management. 80 members and growing. Five events delivered in four months.
The problem
Three co-founders. One community. Multiple channels to maintain — and no technical team.
Running CoLab IA means producing a steady stream of outputs:
- A website that stays current
- Event pages for every workshop
- LinkedIn posts to reach new members
- Newsletters to engage existing ones
- Meeting recaps to keep decisions traceable
Each output has its own format, its own tone, its own structure. And each co-founder has their own way of writing.
graph TD A[3 contributors] --> B[Website updates] A --> C[LinkedIn posts] A --> D[Newsletters] A --> E[Event pages] A --> F[Meeting recaps] B --> G{Different formats<br/>Different tools<br/>Different tone} C --> G D --> G E --> G F --> G G --> H[Inconsistent output<br/>Slow turnaround<br/>Bottlenecks] style H fill:#fde8e8,stroke:#e74c3c
Does this sound familiar?
If you run a small team, a community, or a side project with collaborators — you have likely hit this exact wall. The problem is not skill. It is orchestration.
The traditional solutions do not work at this scale:
- A CMS requires training and a dedicated maintainer
- An agency creates dependency and recurring cost
- “Just use Google Docs” fragments everything — no structure, no consistency, no deployment pipeline
We needed something different. A system where:
- Any co-founder could create content independently
- The output would be consistent regardless of who wrote it
- Publishing would require no technical knowledge
- Any AI tool could assist without custom setup
The solution — an agentic vault
We built a shared workspace in Obsidian — a free, offline, markdown-based editor. But this is not just a folder of notes. It is an agentic vault: a workspace that any AI assistant can read, navigate, and operate within.
What makes it "agentic"?
The workspace contains its own instructions. When you point an AI assistant (Claude, ChatGPT, Cursor, or any other) at the vault, it knows:
- What the team does
- Where files go
- Which template to use
- What tone to write in
- How to publish
No prompt engineering required. The structure is the prompt.
How it works
graph LR A[You describe<br/>what you need] --> B[AGENTS.md<br/>routes to the<br/>right workflow] B --> C[Playbook<br/>guides step<br/>by step] C --> D[Template<br/>enforces<br/>structure] D --> E[Style guide<br/>enforces<br/>tone] E --> F[Finished<br/>content] style B fill:#4a9ede,color:#fff style F fill:#5cb85c,color:#fff
Four layers, each doing one job:
-
AGENTS.md — the routing layer
- A single file that acts as the entry point for every AI interaction
- Maps requests to workflows: “create an event” goes to the event playbook, “write a post” goes to the content playbook
- Works with any AI tool — no vendor lock-in
-
Playbooks — the procedure layer
- Step-by-step workflows for recurring tasks
- The AI follows them like a checklist
- Examples: event creation, event promotion, article publishing, meeting recaps
-
Templates — the structure layer
- Every content type starts from a template
- Templates embed proven frameworks (more on this below)
- No blank pages, no guesswork
-
Style guide — the consistency layer
- Voice, tone, naming conventions, formatting rules
- Applied to every piece of content, by every contributor
- The AI checks against it automatically
One event, three channels
The most powerful workflow: event promotion. One brief generates three outputs — each optimised for its channel.
graph TD A[Event brief<br/>date, topic, audience] --> B[Event page<br/>for the website] A --> C[LinkedIn post<br/>for reach] A --> D[Newsletter<br/>for members] B --> E["AIDA framework<br/>Attention → Interest<br/>→ Desire → Action"] C --> F["PAS framework<br/>Problem → Agitation<br/>→ Solution"] D --> G["BAB framework<br/>Before → After<br/>→ Bridge"] style A fill:#4a9ede,color:#fff
Why different frameworks?
Each channel has different physics. A website visitor is already interested — guide them to sign up (AIDA). A LinkedIn reader is scrolling — stop them with a problem (PAS). A newsletter subscriber trusts you — show them the transformation (BAB).
The templates encode these frameworks. The contributor does not need to know them — the structure does the work.
Same message. Same facts. Adapted for each channel’s audience and format. One playbook handles the entire flow.
The publishing pipeline
Content goes from markdown file to live website without any technical steps for the contributor.
graph LR A[Edit markdown<br/>in Obsidian<br/>or Cursor] --> B[Push to<br/>GitHub] B --> C[GitHub Actions<br/>builds the site<br/>automatically] C --> D[FTP deploys<br/>to Infomaniak] D --> E[Site is live<br/>colabia.ch] style A fill:#fff3cd,stroke:#f0ad4e style E fill:#5cb85c,color:#fff
- Obsidian or Cursor — edit markdown files visually
- GitHub — version control, collaboration, change tracking
- GitHub Actions — automated build on every push
- Quartz — turns markdown into a static HTML/CSS site
- Infomaniak — Swiss hosting, sovereign, privacy-respecting
The result
One of our three co-founders is non-technical. They edit content in Cursor, push to GitHub, and the website updates itself. No CMS. No build commands. No deployment to configure.
It works.
What lives inside the vault
colab-ia/
├── AGENTS.md ← entry point for any AI
├── _ai/
│ ├── playbooks/ ← event, promotion, publishing,
│ │ meeting recaps
│ ├── templates/ ← article, event page, LinkedIn,
│ │ newsletter, recap
│ ├── style/ ← voice, tone, formatting rules
│ └── tools/ ← deployment instructions
├── _private/ ← never published
│ ├── Strategy
│ ├── Governance ← decisions, meeting notes
│ └── Events ← briefs, comms, archives
├── public/ ← becomes the website
│ ├── articles/
│ ├── community/
│ └── resources/
└── .github/workflows/ ← automated build + deploy
Private stays private
The vault enforces a strict boundary. Strategy documents, meeting notes, and draft content live in
_private/and are never published. The AI respects this boundary — it is encoded in the routing rules.
The tools
| Tool | What it does | Cost |
|---|---|---|
| Obsidian | Edit and organise markdown files | Free |
| Cursor | AI-assisted code and content editor | Free tier available |
| GitHub | Version control and collaboration | Free |
| GitHub Actions | Automated build and deployment | Free tier |
| Quartz | Static site generator from markdown | Free, open source |
| Infomaniak | Swiss web hosting | Low cost |
Total infrastructure cost
Hosting only. Every other tool in the stack is free. The entire publishing pipeline — from authoring to deployment — costs less than a single month of most CMS subscriptions.
Skills and concepts mobilised
This project draws on several interconnected concepts. Each one links to a deeper explanation in the learning platform.
- separation-of-concerns — private content and public content live in separate layers, enforced by the system
- human-in-the-loop — the AI proposes, a human confirms. Every workflow includes a confirmation step before acting
- context-engineering — the vault structure itself is the context. Instead of writing long prompts, the file system tells the AI what it needs to know
- iterative-development — templates and playbooks evolve with the team. Each event improves the next workflow
- contracts-and-interfaces — templates act as contracts: mandatory fields, required structure, consistent output regardless of who fills them
What this makes possible
After four months:
- 80 members in the community, growing with each event
- 5 events produced and promoted across three channels
- 3 co-founders publishing independently, without editorial bottlenecks
- 1 non-technical co-founder maintaining the website without technical support
- Weekly site updates since launch
- Zero CMS, zero vendor lock-in, zero recurring tool cost beyond hosting
Could this work for your team?
If you run a small organisation, a community, or a project with multiple contributors — this model scales down to two people and up to any team willing to work in markdown.
The vault structure is open and replicable. The tools are free. The AI layer works with whichever assistant you already use.
Browse the templates to see how the base vault is built.