What specialization means
Outrich combines task-specific prompts, your campaign brief, saved preferences, writing examples, tools and memory. These are runtime instructions and context—not a claim that your account has its own fine-tuned model.
Your voice
Choose a tone, message length and review mode in AI settings. Add examples you have permission to use. Optional sent-mail voice learning creates a private style summary for your account; it is off by default. You can remove saved memories in the Vault.
Training
Private chats, inboxes and campaign records are not automatically exported into a shared training dataset. The repository’s training and evaluation material uses synthetic examples. No fine-tuning job is run just because you save a style preference. Any future training on member content would need a separate, explicit opt-in, a defined purpose and a documented deletion policy.
Model providers and context
The operator selects the inference endpoint and context budget. Models may have large context windows, but retrieval and summaries are bounded, and recall is not guaranteed. Free hosted models have limits; self-hosted inference still needs computing resources. Review the configured provider’s privacy terms before submitting confidential material.
Action controls
Review mode is the default. Automatic sending is available only when the member enables it and campaign requirements are met. Suppression, account access and mailbox limits are enforced outside the model. Research snippets and received messages are treated as untrusted data. Tool use and AI output can still fail, so consequential decisions need human judgment.
Corrections and feedback
Correct the brief, update your writing instructions, remove outdated memories, or open a support request. Model releases should pass the repository’s evaluation suite and controlled provider tests before the operator enables them.