PSYCHE/OS

Digital Psyche Operating System
A local-first pipeline that turns your digital traces into a navigable psychological map.

Experimental, typographic, privacy-aware, and evolving in public.

---

What It Does

PSYCHE/OS reads the data you already produce — chat sessions, bookmarks, code history, browsing traces, notes — and extracts structure from it. Not generic self-help. Not personality quizzes. Computed structure, cross-validated across sources, with evidence you can inspect.

The core idea is simple:

• use data that already describes your behavior
• inspect it locally instead of handing it to opaque systems
• compute structure, not summaries
• move toward a directional vector, not a pile of advice

It does not diagnose. It maps.

---

How It Works

PSYCHE/OS follows a strict principle: keep only what survives cross-source validation. A pattern that shows up in your Claude sessions, your X bookmarks, and your YouTube history is a real signal. One that appears in a single source is noise.

Sources (exported data)
    ↓  source-specific adapters
Extraction (structured signals per source)
    ↓  cross-source synthesis
Patterns (only what survives comparison)
    ↓  dimensional analysis
Map (navigable psychological structure)
    ↓
Outputs (cognitive genome, narrative arc, directional vector)

What the pipeline produces

• Patterns — recurring behaviors cross-validated across sources
• Cognitive primitives — fundamental operations of your thinking
• Archetypes — latent roles and identities across contexts
• Dimensional scores — psychological, cognitive, social, creative, professional, spiritual axes
• Narrative arc — chapters, tensions, current direction
• Directional vector — where your life pattern seems to want to go
• Semantic map — entities, themes, and relationships

The intended end state is not a recommendation engine. It is a directional reading: a vector that summarizes where the strongest signals point when compared side by side.

---

Plug and Play: Feed Your AI Agents

PSYCHE/OS is not just an analysis tool. It generates ready-to-use outputs that plug directly into your AI workflow.

The Integration view in the dashboard produces five export formats, each designed for a different integration surface:

Export	What it does	Target
System Prompt Generator	Full cognitive profile as a system prompt — patterns, archetypes, dimensions, genome, narrative arc. Copy and paste into any AI chat.	Claude.ai, ChatGPT, Gemini, any LLM
MCP Server Configuration	JSON config that exposes PSYCHE/OS as callable tools (get_patterns, get_archetypes, get_dimensions, get_potentials, get_genome, search_analysis).	Claude Code, MCP-compatible agents
Memory Plugin Format	Compact CLAUDE.md block with top patterns, archetypes, genome, and narrative. Drop it into your project config.	Claude Code persistent memory
Structured Data Export	Full analysis as a single downloadable JSON object.	Data pipelines, custom agents, LangChain, CrewAI
API Endpoint Schema	OpenAPI 3.0 spec for serving PSYCHE/OS data via HTTP.	Custom integrations, hosted APIs

In practice: run the pipeline once, open the Integration tab, copy the output you need, and your AI agent has full cognitive context.

The Onboarding view guides you through connecting each data source with copy-pasteable commands for Claude Code, Codex CLI, X/Twitter (via Siftly), YouTube (via Google Takeout), and cloud AI conversations (Claude.ai, ChatGPT, Gemini).

---

Example Output: Cognitive Genome

From 103 documents across 6 sources, the pipeline extracted 8 cognitive primitives from synthetic demo data:

Primitive	Confidence	Evidence
Failure-Driven Learning	0.92	Error logs treated as foundational reference
Systematic Abstraction Descent	0.90	Progressive simplification across frameworks
Fractal Pattern Transfer	0.88	Same structure applied across different substrates
Infrastructure-First Construction	0.88	Scaffolding as a mode of thinking
Empirical-Mystical Oscillation	0.88	Alternating contemplative and technical modes
Cost-Conscious Optimization	0.88	Efficiency as aesthetic principle
Naming-as-Cognition	0.82	Naming as a tool for conceptual transformation
Burst-Process-Burst Rhythm	0.82	Productive silence between output phases

---

What It Maps

The pipeline produces navigable structure across multiple layers:

• Dimensions — six-axis profile (psychological, spiritual, anthropological, social, creative, professional) with five depth levels from surface behavior to archetypal core
• Semantic map — entities, themes, and relationships as a knowledge graph
• Archetypes — dominant, secondary, emergent, and golden shadow figures in tension
• Narrative arc — life chapters, current tension point, possible resolutions
• Patterns and potentials — cross-validated behavioral signals with confidence scores
• Neurodivergence screening — evidence-based, explicitly not diagnosis

---

Supported Sources

Each source has a dedicated SourceAdapter:

• Claude Code — JSONL session histories with conversation extraction and session metadata
• Codex CLI — JSONL session histories with message parsing and CLI version tracking
• X/Twitter bookmarks — Markdown exports (via Siftly) with topic classification and bookmark counts
• YouTube — Markdown playlists and watch history from Google Takeout
• OpenClaw — Local domain knowledge (openclaw-local) and hierarchical memory vault (openclaw-m1)

Additional sources supported through manual prompt handoff:

• Long-form conversations with Claude.ai, ChatGPT, and Gemini
• Adjacent CLI and agent-session workflows that can be normalized into the extraction format

If a source can be exported, it can probably become an adapter. If any of these sound familiar, this repo is probably meant for you:

• a deep archive of Claude or ChatGPT conversations you suspect says more about you than you remember
• Claude Code or Codex sessions that capture how you think while building, debugging, and deciding
• hundreds of X bookmarks that felt important enough to save, but not important enough to ever open again
• a YouTube Watch Later queue that quietly became an accidental map of interests, aspirations, and unfinished lines of inquiry

---

Quick Start

Requirements: Node.js 20+, npm, Python 3.10+ (for vector search helpers)

git clone https://github.com/michelericco/psyche-os.git
cd psyche-os
npm install
npm --prefix web install

Run the interface:

npm --prefix web run dev

Open http://localhost:5173.

Running The Pipeline

Full local flow:

bash scripts/run-full-pipeline.sh

Or run the stages individually:

bash scripts/extract-claude-sessions.sh
bash scripts/extract-codex-sessions.sh
bash scripts/extract-social-traces.sh
bash scripts/synthesize.sh
bash scripts/neurodivergence.sh

Optional semantic search:

pip install chromadb sentence-transformers
python3 scripts/create-embeddings.py
python3 scripts/search-embeddings.py "shadow integration" --top 5

Validation:

npm run validate

---

Repository Structure

• web/: React dashboard
• scripts/: extraction, synthesis, and embedding helpers
• src/: TypeScript core pipeline
  • src/pipeline/adapters/: source-aware adapters (Claude Code, Codex, Twitter, YouTube, OpenClaw)
• tests/: unit, integration, and E2E tests with fixtures
• docs/: methodology, foundations, and deployment notes
• output/: generated analysis artifacts, gitignored except for .gitkeep

Method And Direction

The methodology is intentionally becoming stricter:

• extraction should maximize signal without inflating interpretation
• synthesis should keep only what survives cross-source comparison
• outputs should remain inspectable and evidence-linked
• the final reading should move toward a coherent directional vector

Project docs: Pipeline methodology · Analytic foundations · Prompt evaluation rubric

Design Principles

• Local-first — No cloud, no sync. Your data stays on your machine.
• Evidence-linked — Every insight cites its source. Interpretations are hypotheses, not identity statements.
• Ontology-first — Schema before data. Structure emerges, not imposed.
• Privacy by architecture — Access control is structural, not policy-based.

Privacy And Safety

This project can touch intensely personal material. Please treat it with care.

• Never commit sources/, output/, extraction dumps, chat logs, or generated profiles.
• Never attach real personal datasets to GitHub issues or pull requests.
• Use sanitized fixtures and synthetic screenshots in public discussion.
• Do not treat outputs as diagnosis, therapy, or clinical assessment.
• Treat interpretations as hypotheses that need scrutiny, not identity statements.

The default .gitignore already protects the most sensitive paths.

Contributing

Contributions are welcome, especially in:

• source adapters and importers
• prompt design and extraction normalization
• synthesis logic and evidence calibration
• evaluation, regression fixtures, and quality gates
• scientific grounding across psychology, sociology, anthropology, philosophy, and cognition
• accessibility, performance, and UI refinement

Start with CONTRIBUTING.md. Security: SECURITY.md. Community: CODE_OF_CONDUCT.md.

Project Status

Solid: interface and visual language, local-first baseline, extraction and synthesis surfaces, demo data and documentation hygiene.

Evolving: adapter breadth, synthesis depth, evaluation rigor, directional-vector quality, broader data compatibility.

The project is open now because it already has a clear direction, and it will improve through careful use, critique, and contribution.