What we're shipping.

Extraction engine v1: automatic identification of decisions, commitments, and assumptions from conversation exports, with confidence scoring (0–100%) and human/AI attribution

Agent log ingestion: accept structured action logs from LangGraph, CrewAI, and custom agent frameworks alongside conversation exports

Ghost Preview: see archive stats instantly (conversation count, date range, platform breakdown) while extraction runs in the background

Realtime progress UI: Parsed → Indexed → Extracting → Ready with per-conversation status and estimated time remaining

Idempotent extraction jobs keyed by (workspace, conversation, segment range, extractor version) — re-runs produce identical results without duplicating costs

Client-side unzip: ChatGPT exports (up to 500MB) are parsed entirely in the browser — raw archive data never touches our servers

Selective upload: choose which conversations to import, or import everything. Media references are preserved but deferred by default to control costs

Chunked ingestion pipeline: conversations are processed incrementally with tolerant parsing and automatic schema repair for malformed exports

Claude export support: import Claude conversation history alongside ChatGPT

SHA-256 content hashing on every artifact at creation time — the provenance chain starts at ingestion, not after

Strata Timeline: unified, timeline-native view of events across all imported sources with filtering by platform, actor, and artifact type

Continuity Search: natural language queries across your archive — 'What did I decide about pricing?' returns a cited, evidence-linked answer

Decision Diff: weekly view showing what changed, what reversed, and what commitments went stale

Template library v0: Incident Binder, Weekly Agent Report, and Decision Ledger Snapshot — generate sendable outputs from your timeline

Provenance UX: every extracted artifact shows its source conversation, extraction confidence, and a link to the original context

Demo Archive with 3 personas: platform engineer managing agent incidents, founder tracking investor commitments, ops leader measuring agent ROI

interstrata.ai marketing site: landing page, features, comparison hub, use cases, glossary, articles, and trust documentation

Supabase backend: auth, RLS-protected schema across 8 modules, edge functions for trust event emission

Continuity Ontology v1: 7 core objects (Artifact, Event, Decision, Assumption, Commitment, Actor, Thread) with typed relationships

Custody profile selection at onboarding with Profile B (Assisted E2EE) as default