How we use AI.

Numen is a single agent. It reads what you read — the news the Council writes, the signals on the lattice, the schemas you have adopted, the scenarios you have built. When you ask Numen a question, it composes an answer from those sources and tells you which it drew from. It is not a separate intelligence; it is the layer you ask the lattice through.

Numen runs on Anthropic's Claude family of models. The agent infrastructure, the retrieval, the search, and the conversation memory are Palanor's. The underlying language model is Anthropic's.

Your conversations are private to you. When you talk to Numen, the conversation is stored in your account, encrypted at rest, and visible only to you. We do not share them with other accounts, other organizations, or third parties. The Council does not read them. Other Members do not read them. We do not read them unless required by law or by a security investigation you initiate.

Cross-organization isolation is enforced at the database layer. When you read on behalf of an organization, Numen can only see what your organization has access to. It cannot read another organization's scenarios, signals, contacts, internal data, or conversations. The boundary is not a policy — it is a Postgres row-level security perimeter that the AI itself cannot escape.

We do not train Anthropic's models on your data. Palanor's API contract with Anthropic disables training on the prompts and completions we send. The Council's public Codex posts and your public Futures are published — they are part of the open lattice, so they may appear in third-party search indexes and AI systems that crawl the public web. Everything inside the platform that is not deliberately published stays inside.

Conversation history. Each exchange you have with Numen is stored in your account. You can review past conversations, search them, and delete any conversation at any time. Deletions are permanent within 7 days (this gives us time to honor a deletion request even if our database backups have already rolled).

Telemetry on usage. We keep counts — how often Numen is invoked, how often it answers usefully, how often it fails, the categories of questions stewards ask. We use this to make Numen better. We never tie this telemetry to your handle or your organization in any export that leaves Palanor.

Citations. When Numen answers using retrieved sources, the sources are logged alongside the answer. This means: if Numen ever cites something we later discover to be wrong, we can trace the chain back and fix it.

• Tell you what is happening inside another organization on Palanor.
• Recommend an action solely because a counterparty would benefit from it.
• Generate synthetic signals, fabricated citations, or invented data and present them as real.
• Speak in your name without your hand on the action — Numen drafts, you publish.
• Train on your private conversations or private organizational data.
• Make a clinical, legal, or fiduciary determination on your behalf. Numen reads; you decide.

Language models make confident wrong claims. Numen runs against retrieved sources to keep it grounded, but it can still miscite, misremember, or confidently invent. We treat this as a fact of the medium, not a defect to apologize for. The discipline:

• Every assistant turn carries the citations it drew from. Read them.
• Numen never offers a hard probability without a sourced basis for the number.
• When Numen does not know, we ask it to say so — and we train it to refuse the cheap satisfaction of guessing.

• You can read every conversation you have ever had with Numen, from your settings.
• You can delete any conversation at any time. Deletions are permanent within 7 days.
• You can export your full Numen history as a JSON archive.
• You can turn off conversation persistence — Numen will still answer, but nothing will be stored.
• You can revoke Numen entirely from your account; the panel will no longer appear.

The first three are wired now. The fourth and fifth ship with the Member settings UI when it lands. If you need any of these before then, email privacy@palanor.com and we will take care of it manually.

If Numen says something wrong about your organization, your data, or someone else's data, email trust@palanor.com immediately. We treat trust-and-safety reports as the highest-priority queue we run.