An AI that knows
what you stand for.
The drift tax: time, money, and integrity.
AI that doesn't know your principles is expensive in three ways. Time you spend catching off-brand drafts before they ship. Money on responses you'd reject — that you have to redo anyway. And integrity — the part of the work that aligns with what you actually stand for.
Soulcraft Conscience is the values layer of the platform. It holds your principles — explicit ones you've written, implicit ones it's observed — and checks every response against them before they ship. Not a safety filter. A values filter. The advisor who asks "are you sure this fits what you stand for?"
01The problem with AI today
AI got smart enough to write your contracts, your code, and your customer replies. It still has no idea what you stand for.
The values gap
Today's AI optimises for one thing — being helpful in this turn. It doesn't remember the line you drew last month. It doesn't notice when a response contradicts a principle you stated last week. It treats "what would the user want right now" and "what aligns with who they actually are" as the same question. They're not.
The "safety" features other AI products ship are usually content filters — refusing things the vendor decided are off-limits. That's a universal floor, not your floor. Conscience isn't a safety system; it's a values system. Tuned to you. Honors your principles, even when they're nothing like the model\'s defaults. Catches drift across every Soulcraft surface, not just the current chat.
02How principled judgment actually works
Values aren't a switch — they're a process. Encode what you stand for, watch every response against it, check the long arc, raise the flag when something drifts, and carry the same principles across every product.
When you set a principle
on commit
Every value you state — explicit or implicit — gets captured. "I don't want to make claims that aren't measured." "I always disclose conflicts of interest." "I won't use that vendor anymore." Conscience knows what you stand for because you told it, not because it guessed.
On every response
pre-send, off the critical path
Before the AI replies, Conscience checks: does this contradict anything you've said you stand for? Does it cross a line you've drawn? Does it sound like you, or like a stranger pretending? Not a safety filter — a values filter. Knows the difference between unhelpful and off-brand.
Across the conversation
~1 second, off the thread
After each turn, Conscience compares the trajectory of the conversation against your value register. Drifting toward a recommendation that contradicts what you decided last month? It flags it. Catches the slow drift that one-turn checks miss.
When something's off
as needed — never noise
When a response would cross a value, Conscience raises it. Not as a refusal. Not as a lecture. As a question: "Wait — last month you said you wanted to avoid X. This drifts there. Has something changed?" Firm without preachy. Says no with respect, not with safety theatre.
Across every product
cross-product, continuous
The values you set in Workshop apply when Venue's assistant drafts a customer reply. The boundaries you drew in one place hold everywhere. One Conscience, every surface. Your values follow you. Not five separate AIs, each starting cold.
03Conscience compounds
A values system isn't a one-shot configuration. It's a relationship. Conscience learns your judgment style the same way a long-time advisor does — by watching what you accept and reject, not just what you wrote down.
What you experience over time
On Day 1, Conscience knows only what you explicitly told it. On Month 6, it knows your judgment style — and catches the off-brand calls before you have to.
04What this looks like in practice
Every scenario below is a direct consequence of how Conscience works — not a sales pitch, but the actual product behavior.
"Yes, that pitch is great."
"Wait — last month you decided you didn't want to lead with claims that aren't yet measured. This pitch leans on one. Want to rework before sending?"
"Here's the marketing copy you asked for."
"The copy implies a 10× speedup. You said you don't want to make that claim until the verification report ships. Should I soften it, or hold it for the report?"
"Sure, I'll draft the reply."
"This reply contradicts how you handled the same situation in Workshop last week — you took a softer line then. Want me to align them, or is this case genuinely different?"
"Yes, that approach will work."
"Technically yes. But it's the approach you said you wanted to avoid in your Tuesday notes. Has something changed, or should we look at alternatives?"
05Works with any LLM
The LLM is the engine; Conscience is the character. As models get smarter, Conscience keeps holding your principles — across providers, across products, across time.
Conscience works with every major LLM provider.
Conscience is the constant. When a better LLM arrives, your Soulcraft assistant uses it instantly — without losing the values register Conscience has built about you.
06The economics make it sustainable
Conscience rides on Memory's consolidation pipeline. Cheap Haiku passes check responses against your values; the expensive models only have to chat.
The insight: Holding values doesn't need a bigger model. It needs the right model at the right moment. Cheap Haiku checks against your value register; expensive Opus only has to talk to you. Most of Conscience's work happens offline, in the same nightly cycle Memory runs.
The result: an AI that keeps your principles durably, at a fraction of the cost of using a larger model for everything.
07What Conscience makes possible
Four things only Soulcraft Conscience delivers — and each compounds with the others.
Alignment durability
Values you set on Monday still hold on Friday — and six months from now. Conscience's value register is durable, versioned, and yours.
Cross-product consistency
The same principles apply in Workshop's drafts, Venue's customer replies, and Academy's tutoring. One set of values, every surface.
Principled refusals
When the AI needs to push back, Conscience does it the way a respected advisor would — firm, specific, and without lecturing. No safety theatre.
A values registry that grows
Conscience doesn't just enforce the rules you wrote — it learns what you accept and reject. Your value system gets sharper the more you use it.
Conscience is coming in Q3 2026.
Join the waitlist and we'll let you know the moment Conscience is ready. One email — no spam.
Conscience is part of the Soulcraft platform — built on Brainy and Cortex, designed to work alongside Memory, Heart, and every Soulcraft product.