How the AI actually works here
Foothold Foundry uses AI heavily and deliberately. This page says exactly how: which models, what they see, what they can and cannot do, and the boundaries we hold even when nobody is looking. It exists because "trust us" is not a policy.
The models
The Foundry's tools run on Anthropic's Claude models. We choose the model per tool for quality, not cost theater, and we change models when the quality bar moves. Model use is metered in credits, and every run shows its credit price before you start it. Failed runs are never charged.
The owned-data boundary (the hard rule)
Your owned data, meaning your contact rows, lead lists, and any personal information about the humans in your audience, is never sent to an external AI provider. Not summarized, not sampled, not "anonymized", never.
How that works in practice: when an AI feature drafts outreach, it writes a template with placeholders. Your real contact details merge into that template on our servers, at send time, after the AI's job is over. AI features that research public information send generic public queries, never your data.
And in the other direction: your data is not used to train AI models. Ours, Anthropic's, anyone's.
What the AI cannot do
- It cannot send email. Every send requires a human trigger behind a preview gate, with suppression and one-click unsubscribe enforced.
- It cannot post anywhere on your behalf. Outreach Targets finds where to engage and drafts the angle; the posting is yours.
- It cannot spend without showing you the price first, and it cannot exceed the credits you actually have.
- It cannot silently change your records. Tool outputs land as drafts, reports, or reviewable records that you approve, edit, or reject.
What the AI does do
- Drafts and synthesizes: briefs, personas, competitive maps, pricing recommendations, requirements, decision records, copy, and plans, grounded in your venture's own project memory.
- Pushes back: the Definition gut check argues with weak framing instead of applauding it, and the Validation Scorecard says "not yet" when the evidence says so.
- Shows its work: reports state assumptions, methods, and what to verify, so you can check the reasoning instead of trusting the tone.
If this page and reality ever disagree
Tell us: notifications@footholdfoundry.com. This page is a commitment, not marketing, and we would rather fix a violation than explain one.