DiffMem in production, Git-based AI memory
Git worktrees as memory backend for WhatsApp companion—genuinely novel approach to AI continuity.

Memory-aware spiritual AI with 21% conversion, but astrology apps already saturated.
Spiritually-oriented individuals seeking personalized daily reflection and guidance
Co-Star · AstroTalk · Replika
We’ve been quietly building and testing nummi.ai, a personal AI companion that:
Remembers your journey and context over time,
Helps you reflect, understand yourself, and navigate life with clarity,
Offers daily guidance & personalized insights (including spiritual/astrology-based timing) based on your birth chart and tracked patterns.
Unlike generic chat assistants, Nummi is designed as an ongoing companion — not a search tool or generic productivity bot. It learns from your inputs, your patterns, and your preferences so responses get more contextual and relevant over time.
Why we built it
In early testing:
We’ve seen 10k+ signups
(~21% signup-to-paid conversion) These are very early numbers, but users seem to value the ongoing connection and personal context. (This isn’t “astrology predictions” — it’s reflective AI that combines memory + optional vedic insights.)
We’re experimenting with features like:
Personalized daily insight prompts
Habit support and emotional check-ins
Memory that grows the quality of responses over time All with privacy and optionality at the core.
Open questions for HN
How can we better signal long-term value (vs one-off chat)?
What onboarding patterns help users unlock value faster?
To product people: how do you think about AI companions that mix reflection + memory + optional metaphysical context?
Happy to answer any questions — and genuinely looking for feedback.
Git worktrees as memory backend for WhatsApp companion—genuinely novel approach to AI continuity.
Cross-project memory for coding agents, but MCP ecosystem is nascent and fragmentation risk high.
Scrappy v1 with dangerous setup flags and zero stars on GitHub.
Replika clone with persistent memory, but the market is already saturated.
Mr. Relentless actually pushes back instead of endlessly validating — refreshing in AI wellness.
Mimir hooks into Claude Code lifecycle events so agents can 'mark' facts (e.g., "API uses snake_case") into a DuckDB-backed memory and RAG pipeline, then auto-injects that context as additionalContext for later agents. It's a pragmatic, well-scoped solution to the annoying problem of agent amnesia — very useful if you run agent swarms, but its impact is limited by Claude Code adoption and the need for the surrounding infra (BGE keys, hooks).