AI & Chat Data Policy

What This Covers

This policy explains how data is handled when you use Talk to EWE (our AI assistant), the AI Council, Berkeley gamification, the Seeping Fog system, and any other AI-powered feature across the EWEPIP ecosystem.

This policy supplements our Privacy Policy. If there is a conflict, the more protective provision applies.

What We Collect During AI Chat

How Chat Data Is Used

The data collected during your AI conversations serves these specific purposes and no others:

• Contextual responses — Your messages within a session are used to provide relevant, context-aware responses. EWE remembers what you said earlier in the conversation so she does not repeat herself or ask you the same question twice.
• Cognitive level progression — Your cognitive level (L1 through L5) is maintained across sessions so EWE can adapt her communication style. A returning expert user should not receive beginner-level explanations.
• AI personality calibration — Aggregated, anonymized interaction patterns (not individual conversations) help us improve how the four AI personalities respond. For example, if 80% of users find a particular response confusing, we adjust the response template.
• Error debugging — When the AI behaves unexpectedly (wrong personality, incorrect tool execution, nonsensical response), session metadata helps our engineering team identify and fix the issue.

How Chat Data Is Stored

• Session context — Stored in Redis. Expires automatically after 24 hours of inactivity. When you close your browser and come back the next day, your previous session context is gone unless you opted into history saving.
• Cognitive level — Stored in PostgreSQL. Persists across sessions and is tied to your account. This is a single number (1-5) and a few metadata fields, not your conversation content.
• Chat history — Not permanently stored by default. If you want to keep your conversations, enable "Save History" in your account settings. Saved conversations are stored in PostgreSQL and can be deleted at any time.
• Anonymized interaction patterns — Retained for system improvement. These are statistical aggregates (e.g., "47% of route planning requests come from the drivers sector") with no personally identifiable information.

AI Training

To be very specific:

• Talk to EWE uses Qwen models running on Ollama, hosted on our own physical GPU servers (Dell R720 with Tesla P40 GPUs).
• Your chat messages are sent to our server, processed by our local AI model, and the response is sent back. At no point does your message leave our infrastructure to reach a cloud AI provider.
• We do not fine-tune or retrain AI models using individual user conversations.
• If we ever add the option to use cloud AI providers (for features requiring capabilities beyond our local models), we will update this policy and require explicit opt-in consent before routing any of your data to external services.

Your Controls

You have full control over your AI interaction data:

• Delete chat history — Account Settings → AI → Clear History. This permanently removes all saved conversations.
• Disable cognitive level tracking — Account Settings → AI → Disable Cognitive Tracking. EWE will treat every session as if you are a new user (default L3 behavior).
• Use AI anonymously — On talentliberation.org, many AI-powered tools work without any login. No account means no persistent data.
• Request full AI data deletion — Email [email protected] to request complete deletion of all AI interaction data associated with your account. We will process the request within 30 days.
• Export your data — Account Settings → Data → Export. Download all data we hold about you, including any saved chat history, in a machine-readable format.

Seeping Fog Data

The Seeping Fog system is part of the EWEPIP experience layer. It tracks a small set of behavioral metrics to adapt the visual presentation of the site:

• What it tracks — Visit count, message count, and which domains in the ecosystem you have explored. These are simple counters.
• What it does not track — Message content, page-level browsing patterns, click paths, or any form of detailed behavioral profiling.
• Your control — You can reset your fog level at any time in Account Settings → Experience → Reset Fog Level. This sets all counters back to zero.

Gamification Data

The Berkeley gamification system tracks experience points (XP), achievements, daily login streaks, and domain exploration:

• XP and achievements — Tied to your account and stored in PostgreSQL. Earned through genuine platform interaction.
• Leaderboards are opt-in — Your gamification data is visible only to you by default. You must explicitly opt in to appear on community leaderboards.
• Your control — Account Settings → Experience → Hide from Leaderboards. You can also request a full gamification data reset, which clears all XP, achievements, and streaks.

AI Council

When you ask a question that triggers the AI Council (multiple personalities weighing in on a decision), the deliberation process works as follows:

• Your question is processed by multiple AI personality configurations (EWE, BERKELEY, RAM, BUBO) in parallel.
• Each personality generates its perspective independently.
• The council vote and reasoning are generated in real-time and returned to you as part of the conversation.
• Council deliberations are not stored separately from your normal conversation data. The same storage and deletion policies apply.

Third-Party AI

We currently use Ollama with Qwen models running entirely on our own hardware. Our commitment:

• If we ever integrate third-party AI APIs (OpenAI, Anthropic, Google, etc.), we will update this policy at least 30 days before activation.
• Third-party AI will be opt-in, not opt-out. You will need to explicitly agree before any of your data is sent to an external AI provider.
• We will clearly label which features use local AI versus third-party AI so you can make informed choices.
• We will never silently switch from local to cloud AI processing.

AI Teaching & Transparency

For details on how our AI teaches, grows with you, and progressively discloses truth based on your readiness, see our dedicated AI Transparency & Teaching Policy. It covers:

• The Anti-MCI (Mass Cognitive Incapacitation) philosophy
• 4 teaching modes (Socratic, Truth, Growth, Agentic)
• 5 progressive disclosure levels
• How information flows through corridors
• The skill tree and XP system

Contact

For questions about AI data handling, to exercise your data rights, or to report concerns:

• AI data inquiries: [email protected]
• General privacy: [email protected]

EWEPIP is operated by Salmagundi Services LLC.