Ateliersavant In-House
GHOST — Relationship Intelligence Without Inference
Every relationship management tool on the market tells you how to feel about people. Ghost tells you what actually happened.
Every CRM Lies to You
The relationship tools that exist — CRMs, contact managers, AI assistants — share a fundamental design flaw: they infer. They score your relationships. They decide who is important, who is drifting, how warm a conversation is, whether you should be worried. They wrap these verdicts in probability language and confidence scores, but they are still verdicts — generated from your private data, without your consent, without auditability, and without recourse.
The result is a system that tells you how to feel about people based on pattern-matching your most intimate conversations. That is not intelligence. It is surveillance dressed as productivity.
Ghost was built to answer a different question: what if a relationship intelligence system only told you what actually happened — and left every interpretation to you?
The interface is in development.
Ghost is currently in active development — the ingestion engine, identity resolution layer, and audit architecture are complete. The iOS and Mac applications ship when they are ready, not before.
Raw Truth is Immutable
Ghost is built on a single structural principle: raw conversation data, once imported, is never modified. Every interpretation — identity resolution, relationship mapping, pattern recognition — exists as a separate, reversible, auditable layer on top of the immutable record.
ObservedIdentity
Written once from raw import data. Never modified. The source of truth.
ResolvedIdentity
Interpretable, reversible, auditable. Every change produces an audit entry.
IdentityAuditEntry
Every interpreted operation is logged. Full state captured. Replay is always possible.
"No psychological analysis. No emotional inference. No relationship quality scoring. No trust scores. No tone analysis. No behavioral prediction."
— Ghost Hard Constraints, enforced at the schema level
Structural Analysis.
Nothing More.
Ghost ingests real conversation data from WhatsApp, Telegram, email, and pasted chat — and builds a structural map of who you communicate with, how often, and in what context. It resolves identities across platforms. It assembles relationship dossiers from verified facts. It does nothing else.
WhatsApp, Telegram, email, pasted chat
Detection, code-switching, role assignment
Cross-platform resolution, deduplication
Verified facts only, fully auditable
All identity resolution is deterministic and rule-based. No LLMs in the ingestion layer. No embeddings. No vector search. The same input always produces the same output.
10 Skill Packages.
One Doctrine.
ghost-ingestion
Multi-platform conversation parsing, zero runtime dependencies
ghost-language
Language detection, code-switching, participant role assignment
ghost-relationship
ObservedIdentity, ResolvedIdentity, dossier assembly, audit
ghost-persistence
Supabase persistence layer, user confirmation flows
Tests Passing
Runtime Dependencies
Deterministic
The constraint is the product.
In a market where every AI relationship tool races to analyze and score personal data, the tool that explicitly refuses to do so occupies an uncontested position. Ghost is not a less capable CRM. It is a fundamentally different kind of system — one that treats your relationships as yours, not as training data.
Next Case Study