An AI relationship co-pilot built on Claude Code. It learns about you and your partner and generates experiences that feel like they came from someone who's known you for years. Not a generic recommendation engine. Not a therapist. Just a system that gets you.
Everything runs locally. No app, no server, no subscription.
| Command | What you get |
|---|---|
/onboard |
5-minute setup interview, then your first date idea on the spot |
/date-idea |
One date suggestion + 2 alternatives for this week |
/date-batch |
Four dates for the month ahead (engineered for variety) |
/trip-idea |
Destination + what to do there (no hotels or flights) |
/activity-pack |
Five shared hobbies, rituals, and projects to try |
/checkin |
Five conversation starters tailored to your relationship stage |
/week-ritual |
One recurring weekly ritual to build into your routine |
/remember |
Update your profile anytime ("we loved that," "we're moving") |
/feedback |
Quick 3-question debrief after any date, trip, or activity |
Every suggestion includes a title, time estimate, cost estimate, and a "why it fits you" section that references your actual context. If a suggestion could apply to any random couple, the system rewrites it.
- Node.js 18+
- A Claude.ai account (any plan)
# Install Claude Code
npm install -g @anthropic-ai/claude-code
# Clone and enter the project
git clone https://github.com/tanmayiift/couple-os.git
cd couple-os
# Launch
claude/onboard
That's it. Three quick rounds of conversation (who you are, what you like, practical stuff), and then the system generates your first personalized date idea right there. Takes about 5 minutes.
After that, run /date-idea anytime you need a plan.
Couple OS starts useful and gets better the more you use it.
Active feedback:
- Run
/feedbackafter any date or trip for a quick 3-question debrief - Run
/rememberanytime to tell it something new ("partner hates loud restaurants," "we want to try kayaking")
Passive learning:
- During any conversation, if you mention a preference or reaction, it'll ask: "Want me to note that?"
- After every suggestion it saves, it appends a prompt: "Done this? Tell me how it went."
What it tracks (all local, all yours):
memory/past-dates.md— dates tried + ratings (avoids repeats)memory/past-trips.md— trips taken (avoids repeat destinations)memory/learnings.md— discovered preferences ("prefers small venues," "best dates are unplanned")memory/wishlist.md— things you've said you want to try
The richer these files get, the more specific every suggestion becomes.
couple-os/
├── CLAUDE.md # System brain — read every session
├── .claude/skills/ # Slash commands
│ ├── onboard/SKILL.md
│ ├── date-idea/SKILL.md
│ ├── date-batch/SKILL.md
│ ├── trip-idea/SKILL.md
│ ├── activity-pack/SKILL.md
│ ├── checkin/SKILL.md
│ ├── remember/SKILL.md
│ ├── feedback/SKILL.md
│ └── week-ritual/SKILL.md
├── context-library/ # Your couple profile (gitignored)
│ ├── couple-profile.md
│ ├── preferences.md
│ ├── financials.md
│ ├── constraints.md
│ ├── local-context.md
│ └── memory/
│ ├── past-dates.md
│ ├── past-trips.md
│ ├── learnings.md
│ └── wishlist.md
├── context-library/*.template.md # Blank templates (shipped with repo)
├── sub-agents/quality-rules.md # Output quality enforcement
├── outputs/ # Generated suggestions
│ ├── dates/
│ ├── trips/
│ ├── activities/
│ └── check-ins/
└── setup/
├── installation-guide.md
└── first-session-checklist.md
All data stays on your machine. The context files (your names, budget, preferences) are plain markdown in this directory. They're gitignored by default so they never get committed.
Nothing is sent to any server except your prompts to Claude via the standard Claude Code API (same as any Claude conversation).
You can use aliases instead of real names. You can skip any question during onboarding. More detail means better suggestions, but it's always your call.
This is a template. Everything is meant to be changed:
- Edit
context-library/files directly anytime - Add your own slash commands to
.claude/skills/ - Modify output formats in any SKILL.md
- Adjust quality rules in
sub-agents/quality-rules.md
Three conversational rounds:
- Who are you two? Names, city, relationship stage, living situation
- What makes you tick? Interests, food, energy level, wishlist, hard nos
- Logistics and money Budget, schedules, transport, local favorites
Geography (nearby cities, airports, climate) is auto-populated from your city. You only answer things only you would know.
Ends by generating your first /date-idea so you see the payoff immediately.
Flags: low-effort, splurge, outdoor, or any weekday name.
Returns 1 primary suggestion with full detail + 2 alternatives at different effort levels. References your preferences, budget, and location. Checks past dates to avoid repeats.
Four dates for the month. Engineered for variety:
- At least 1 indoor + 1 outdoor
- At least 1 free + 1 paid
- At least 1 weeknight-friendly
- No two in the same category
Flags: weekend, week, long-haul, any season, or options (for 3 brief picks).
Deep-dive format: what to do there, day structure, food culture, activity budget, best time to visit, and why it fits you specifically. No hotel or flight recommendations.
Five non-date activities: new hobbies, recurring rituals, creative projects, physical anchors. Mix of one-time and recurring. Each includes a specific first step so you can start this week.
Five open-ended questions calibrated to your relationship stage. Not therapy. Variants: after-date, after-trip, monthly (10 prompts).
One small recurring habit: anchored to a specific time, sustainable, and specific to your life. Appended to a running list so you build a collection over time.
Three questions: What was it? How did it go? One thing to remember? Updates memory files automatically. Under 2 minutes. This is what makes future suggestions smarter.
Natural language updates to any context file. "We loved the hiking date." "Partner gets anxious in crowds." "We're moving to Berlin in October." It routes to the right file and confirms what changed.
PRs welcome. Especially interested in:
- New slash commands (anniversary planner, seasonal bucket list, etc.)
- Better output formats
- Non-English locale support
- Quality-of-life improvements
MIT — use it, fork it, make it yours.