The Open-Source AI Productivity Series for EO Operators. Your Pain → Our Skill → Continuous.
A continuous, audience-driven series where EO members surface real AI-setup pains, and the community ships verified solutions. Hosted by Christoph Erler, Dominik Raute (CTO JustWatch), and Fabian Gless, EO Berlin.
flowchart LR
Brain([Build a personal AI brain<br/>that actually works])
Brain --> S1[<b>Setup 1, Chris</b><br/>Workspace-Native<br/>━━━━━━━━━━━<br/>~150 EUR/mo all-in<br/>Mobile via Telegram<br/>~23h/wk saved<br/>Value in 60 min]
Brain --> S2[<b>Setup 2, Dom</b><br/>Local-First Rolodex<br/>━━━━━━━━━━━<br/>$10-50/mo + Mac hardware<br/>5-tier memory · 107 Rolodex<br/>Sovereignty · No mobile<br/>Backbone in 90 min · Full in 12 months]
Brain --> S3[<b>Setup 3, Fabian</b><br/>PAI Life OS<br/>━━━━━━━━━━━<br/>$20-100/mo + Mac hardware<br/>45 skills · 171 workflows · 37 hooks<br/>Pulse dashboard · macOS/Linux<br/>Value in 1-2 days]
classDef chris fill:#dbeafe,stroke:#2563eb,stroke-width:2px,color:#0f172a
classDef dom fill:#dcfce7,stroke:#16a34a,stroke-width:2px,color:#0f172a
classDef fabian fill:#f3e8ff,stroke:#9333ea,stroke-width:2px,color:#0f172a
classDef start fill:#fef3c7,stroke:#d97706,stroke-width:3px,color:#0f172a
class Brain start
class S1 chris
class S2 dom
class S3 fabian
TL;DR, for ~80% of EO operators, the answer is Setup 1 (Workspace-Native, Chris). You already live in Workspace, you want value in 30 days not 12 months, you want mobile, and you can afford ~150 EUR/mo. The other two are for specific edge cases (regulated data → Dom, opinionated Life OS lover on Mac → Fabian).
The honest one-question filter:
flowchart TD
Start([Where do you spend your day?])
Start --> Q1{Inside Google Workspace<br/>Gmail, Calendar, Drive?}
Q1 -- Yes --> Q2{Can you spend<br/>~150 EUR/month?}
Q1 -- No --> Q3{Do you handle data<br/>that cannot leave your machine?}
Q2 -- Yes --> S1[<b>SETUP 1, Chris</b><br/>Workspace-Native<br/>~150 EUR/mo · ~23h/wk saved<br/>Mobile via Telegram bot]
Q2 -- No --> Q3
Q3 -- Yes --> S2[<b>SETUP 2, Dom</b><br/>Local-First Rolodex<br/>~$0/mo marginal · Mac M-series<br/>You debug your own stack]
Q3 -- No --> Q4{Do you want a complete<br/>Life OS out of the box?}
Q4 -- Yes --> S3[<b>SETUP 3, Fabian</b><br/>PAI Life OS<br/>Free + your API spend · Mac/Linux<br/>45 skills · 171 workflows · 37 hooks]
Q4 -- No --> S1
classDef chris fill:#dbeafe,stroke:#2563eb,stroke-width:2px,color:#0f172a
classDef dom fill:#dcfce7,stroke:#16a34a,stroke-width:2px,color:#0f172a
classDef fabian fill:#f3e8ff,stroke:#9333ea,stroke-width:2px,color:#0f172a
classDef question fill:#fef3c7,stroke:#d97706,stroke-width:2px,color:#0f172a
class S1 chris
class S2 dom
class S3 fabian
class Q1,Q2,Q3,Q4 question
Why Setup 1 is the default for the EO audience: of 118 audience registrations across 53 chapters at Event #1, 69% already use Claude and the dominant pain cluster was "Setup itself / time to configure". The Workspace-Native path is the lowest-friction path to a working brain in <60 minutes. The other two are honest counterfactuals: Dom's local-first proves you can do this at near-zero marginal cost if you have 12 months of engineering time and a Mac, Fabian's PAI proves you can skip the build entirely if you accept Miessler's opinionated framework + a single Claude Max (or equivalent) subscription.
Three operator-level setups demoed at Event #1 (2026-05-11). Same problem, build a personal AI brain that actually works in production. Three different operator-level answers. Pick the one closest to how you already work, fork it tonight, run it tomorrow.
| Setup 1, Workspace-Native | Setup 2, Local-First Rolodex | Setup 3, PAI Life OS | |
|---|---|---|---|
| Operator | Christoph Erler (EO Berlin) | Dominik Raute (CTO JustWatch) | Fabian Gless (EO Berlin) runs Daniel Miessler's PAI |
| Stack | Claude Code + 31 MCP + 19 skills + 125 memory + 7 crons + Telegram | Vanilla OpenClaw + custom 5-tier memory + Rolodex + local LLMs on Mac | PAI v5.0.0: 45 skills + 171 workflows + 37 hooks + Pulse dashboard |
| Cost / month (honest, steady-state) | ~150 EUR all-in (Claude Pro 20 + AWS Lightsail 20 + APIs ~110) | $10-50/mo cloud Claude escalations + electricity (NOT $0, see below) | $100-200/mo single Claude Max (or equivalent Codex / Gemini) subscription. PAI is agent-agnostic. |
| Hardware up-front | Whatever you already have (Mac/Linux/Windows+WSL) | Mac M-series + 32GB+ RAM ≈ 3-5k EUR once if you don't have one | Anything that runs your chosen agent's CLI. Mac or Linux. No special hardware tier. |
| 3-year TCO (no new hardware needed) | ~5.4k EUR | ~$360-1.800 | ~$3.600-7.200 |
| 3-year TCO (incl. ~$4k Mac amortized) | ~5.4k EUR | ~$4.360-5.800 | ~$3.600-7.200 (no new hardware needed) |
| Mobile? | ✅ Telegram bot | ❌ Desktop-first | Dashboard at localhost:31337 |
| Platform | Mac / Linux / Windows+WSL | Mac M-series (Linux possible, theoretical) | macOS / Linux only |
| Tracked time saved | ~23h/wk | similar order, $0 marginal cost | tied to PAI's 7-phase Algorithm |
| Time to first working skill | 60 min | 90 min for backbone, 12 months for full Rolodex | 30 min install + 1-2 days configuration |
| Pick if | You live in Gmail/Calendar/Drive all day | You handle data that cannot leave your machine | You want opinionated framework, not blank slate |
| Skip if | You need full local sovereignty | You want value in 30 min | You're on Windows |
| Detailed spec | → setups/chris-claude-code.md | → setups/dom-rolodex.md | → setups/fabian-personal-ai.md |
The honest cost insight. "$0 marginal" or "free" headlines hide hardware + escalation API costs. Once you amortize a Mac and add cloud Claude calls for hard reasoning, all three setups land in roughly the same 3-year TCO bracket for someone who needs to buy hardware. The real differentiator is time-to-first-value (60 min vs 1-2 days vs 12 months) and what you optimize for (mobile vs sovereignty vs opinionated framework).
🔵 Setup 1, Chris (Workspace-Native). The "AI as infrastructure replaces a team I would have hired" path. Claude Code wired into Google Workspace via 31 MCP servers, 19 reusable skills (8 of them in /skills/), 125 memory files routed by a ~50-line CLAUDE.md, mobile via a Telegram bot on AWS Lightsail Frankfurt. ~23h/wk saved on a tracked 8-week sample. Replaces ~1.5 FTE (80-110k EUR/yr) at a 45-60x cost ratio. Personal opportunity cost reclaimed: ~270k EUR/yr. → Read the full spec, origin, 4-layer brain, 8 forkable skills with names, 7 cron jobs, anti-AI voice rules, 30-day rollout plan, troubleshooting.
🟢 Setup 2, Dom (Local-First Rolodex). The "sovereignty over convenience" path. Vanilla OpenClaw as backbone, custom 5-tier memory (System Prompt > Bootstrap > On-Demand > Search Index > Raw Archive), a Rolodex of 107 person dossiers, 1.183 files indexed across 5.735 vectors / 11 collections, all running on local models on a Mac. 2-5 second queries. Real cost: $10-50/mo cloud Claude escalations + electricity, plus Mac M-series hardware (~3-5k EUR once if you don't have one). Built over 12 months from scratch. The agent that remembers. → Read the full spec, 5-tier architecture, what Dom shared at the event, honest struggles, when this is right (and when it is not).
🟣 Setup 3, Fabian (PAI Life OS). The "complete opinionated Life OS out of the box" path. Daniel Miessler's PAI v5.0.0 (12.100+ stars, MIT), 45 skills, 171 workflows, 37 hooks, a Pulse dashboard at localhost:31337, the Telos / ISA / DA conceptual framework. One-line install. macOS or Linux. Real cost: PAI itself is free + MIT, plus one single Claude Max (or equivalent Codex / Gemini) subscription at ~$100-200/mo. PAI is agent-agnostic, no special hardware tier needed. → Read the full spec, what you get, the Telos step that matters, honest struggles, why this matters as a reference architecture.
Brand new to GitHub or AI tooling? Start at
START-HERE.mdfor the 3-step path (fork → clone → first skill). No jargon, screenshots from official GitHub Docs.
Don't know what Claude Code, MCP, or Skills are? Plain-English glossary at
resources/glossary.md.
Looking for a fix to a specific pain?
SOLUTIONS.md, 11 documented solutions, one per audience pain cluster from 118 registrations.
A living solution-pipeline for the AI-productivity pains that founders and operators actually face. Every event ends with a git push, not just „thanks for listening". Every audience pain becomes either a documented solution, a working skill, or both.
Three pillars:
- Audience-driven curriculum. Pains come from registration data and live Slido, not speaker egos.
- Open-source solutions. Every challenge gets a skill / template / playbook here.
- Continuous cadence. 30-90 days between events, depending on market velocity and community demand.
11 in-depth solutions covering all 10 pain clusters identified from 118 audience registrations across 53 EO chapters and 4 continents. See SOLUTIONS.md for the full index and AUDIENCE-ANALYSIS for the verified pain breakdown.
- Joint deck · chris1928a.github.io/eo-ai-exchange/events/01-2026-05-11-setup-trap/slides.html (22 slides)
- Chris's deep dive · chris-demo.html (44 slides)
- Q&A backup · qa-deck.html (15 slides, all 12 Slido questions with sources)
- Setup Trap Diagnostic, 1-pager · the 3 questions to escape the trap
- 30-Min aiOS Blueprint · zero to working setup in 30 min
- OpenClaw Honest Assessment · incl. CVE-2026-25253 + 5 alternatives
- GDPR Claude Checklist DACH · incl. EU AI Act August 2026 deadline
- MCP Cookbook · 10 essential MCP servers for founders
- Team Rollout Playbook · 90-day plan with BBVA case study
- Plus 5 more, one per pain cluster, in
solutions/
External open-source patterns we recommend forking, reading, or learning from.
github.com/danielmiessler/Personal_AI_Infrastructure · MIT · 12.100+ stars
A Life Operating System for AI. PAI captures who you are, what you care about, and where you are trying to go, and then helps you move toward it. Three layers:
- PAI itself, the OS (skills, memory, the Algorithm, your Telos, identity files)
- Pulse, the Life Dashboard at
localhost:31337 - The DA (Digital Assistant), the voice you talk to
v5.0.0 ships 45 skills, 171 workflows, 37 hooks, Algorithm v6.3.0 (Current State → Ideal State across seven phases), the ISA primitive (universal "ideal state" articulation), and structural privacy via containment zones. macOS and Linux supported (Windows not yet). One-line install: curl -sSL https://ourpai.ai/install.sh | bash.
Why we reference it: PAI is the opinionated counterpoint to Chris's workspace-native setup and Dom's local-first memory. It is featured as Demo 3 at Event #1, presented by Fabian Gless. Companion deep dive: events/01-2026-05-11-setup-trap/fabian-demo-pai.md.
eo-ai-exchange/
├── README.md ← you are here
├── START-HERE.md ← beginner's 3-step path (fork → clone → first skill)
├── SOLUTIONS.md ← master index of all solutions, by pain cluster
├── CONTRIBUTING.md ← how to PR your own skill or solution
│
├── events/ ← one folder per event
│ └── 01-2026-05-11-setup-trap/
│ ├── README.md ← event-specific intro
│ ├── slides.html ← joint deck for the live event (22 slides)
│ ├── chris-demo.html ← Chris's solo deep dive (44 slides)
│ ├── qa-deck.html ← Q&A backup deck (15 slides)
│ ├── fabian-demo-pai.md ← PAI demo companion (1200 words)
│ ├── QA-CHEATSHEET.md ← panel cheatsheet, plain Markdown
│ └── AUDIENCE-ANALYSIS.md ← anonymized pain breakdown, 118 registrants
│
├── setups/ ← three operator setups, fork-ready
│ ├── chris-claude-code.md ← Workspace-native, ~150 EUR/mo, 23h/wk saved
│ ├── dom-rolodex.md ← Local-first 5-tier + Rolodex, ~$0/mo
│ └── fabian-personal-ai.md ← Daniel Miessler's PAI v5, complete Life OS
│
├── skills/ ← 8 forkable starter SKILL.md files from Event #1
│ ├── morning-brief/ ← Daily 7am brief (2.25h/wk saved)
│ ├── diarize-person/ ← 1-page stakeholder dossier
│ ├── draft-by-channel/ ← Voice-locked drafts per channel (2.7h/wk)
│ ├── weekly-review/ ← Friday 7-day review (90→10 min)
│ ├── memory-curator/ ← Weekly memory hygiene cron
│ ├── audit-process/ ← Domain example: process diagnostics
│ ├── sales-script-rewriter/ ← Domain example: sales call coaching
│ └── property-pricing/ ← Domain example: daily revenue management
│
├── templates/ ← starter foundation files
│ ├── CLAUDE.md.template ← Project-level instructions, ~50 lines
│ └── memory-templates/ ← user_about, feedback_voice, hat, curator rules
│
├── solutions/ ← solutions grouped by pain cluster (10 clusters)
│ ├── tool-overload/ ← Tool Overload Diagnostic
│ ├── setup-itself/ ← Setup Trap Diagnostic + 30-Min aiOS Blueprint
│ ├── agents/ ← Agent Reliability Checklist
│ ├── team-adoption/ ← Team Rollout Playbook (90-day)
│ ├── integration-mcp/ ← MCP Cookbook (10 essential servers)
│ ├── openclaw-honest/ ← OpenClaw Honest Assessment (incl CVE)
│ ├── industry-healthcare/ ← Healthcare Scheduling Stack
│ ├── security-gdpr/ ← GDPR Claude Checklist DACH
│ ├── pace-keeping-up/ ← Weekly AI Filter
│ └── time-learning/ ← 60-Day Founder Onboarding
│
├── resources/
│ └── glossary.md ← plain-language defs (Claude Code, MCP, Skills, etc.)
│
└── speakers/ ← speaker pipeline + alumni
Cadence over Quarterly. Solutions over Slides.
Events fire on triggers, not on a calendar:
- Major model release → next event in 30 days, focused on what changed
- Major tool / MCP standard shift → next event in 30 days, focused on what to build
- Audience pain cluster maturing (10+ submissions on a new topic) → next event in 60 days, industry/vertical spotlight
- Default cadence (no trigger) → every 90 days, solution review + new cycle
Hosts check triggers monthly. The next event is announced when there is something genuinely worth meeting for.
Three ways to contribute:
- Submit a pain. Open an issue tagged
paindescribing what you are stuck on. We aggregate these for the next event. - Submit a solution. Open a PR with a markdown file or a runnable skill in the appropriate folder. See CONTRIBUTING.md for format.
- Apply to speak at the next event. See speakers/README.md.
All contributors are credited in speakers/alumni/.
All event content is shared under Chatham House Rule: share the information, not the attribution. This applies to the GitHub repo too: if you reference a discussion from an event, do not cite specific speakers without their explicit permission.
- Christoph Erler (EO Berlin), former co-founder & COO at ComX (exit). Founder of Erler Ventures.
- Dominik Raute (EO Berlin), CTO of JustWatch.
- Fabian Gless (EO Berlin), founder of Die Tierversicherer (AI-enabled insurance broker, performance marketing roots). Personal AI Infrastructure (PAI) operator.
All solutions in this repo are released under MIT License. Use, modify, share. Attribution appreciated, not required.
Last updated: 2026-05-14 (Event #1 wrapped, 11 solutions + 3 setups + glossary live, 118 registrations across 53 EO chapters).