Source-grounded advisory infrastructure
A council for the mission that matters now.
Not a prompt asking AI to impersonate famous people. Each mentor is grounded in their own documented corpus, cites what informed the answer, and joins a council that turns distinct counsel into a plan the user approves.
The problem with AI mentors
A prompt changes the performance. A corpus changes the evidence.
A general language model predicts a response from patterns learned during training and the context in the current conversation. Naming a mentor can steer its tone and point of view. It does not, by itself, make the model open that mentor's books, search a defined edition, stay inside that work, or show what actually supports the advice.
A prompt tells the AI how to answer. A retrieved corpus gives it evidence to answer from.
It can sound right and still be ungrounded.
“Act as this famous thinker” is a role instruction, not research. The model may recall useful ideas, but it can also blend the mentor with other writers, modern commentary, or a plausible guess—all in the same confident voice.
- No exact source boundary
- No guarantee a relevant passage was consulted
- No clean line between the mentor and the model
The source material changes the answer.
Before a response is composed, Imaginal AI retrieves relevant passages from that mentor's separately governed corpus. The model reasons with those passages in context, cites them, and distinguishes source evidence from present-day application.
- One governed corpus for each mentor
- Relevant passages retrieved before generation
- Visible citations and honest evidence gaps
One corpus per mentor. One plan across the room.
Each council seat retrieves independently, so Marcus Aurelius does not quietly answer from Machiavelli. Only after the distinct perspectives are visible does the Meta-Advisor map agreements, tensions, unknowns, and a plan for the user to approve.
Inspect how source grounding works →
How the system works
Different minds should remain different.
Name the mission
Describe the objective, the current constraint, the stakes, and what a useful decision would change. The mission—not a famous name—determines which capabilities belong in the room.
Assign the seats
Choose mentors by function: strategy, incentives, operations, persuasion, identity, resilience, ethics, or another needed discipline. Every seat receives an explicit role.
Retrieve separately
Each mentor searches only its own governed corpus. Passages, locators, and source metadata stay attached to that mentor instead of dissolving into one blended persona.
Commit and review
The Meta-Advisor surfaces disagreement, records a decision, and turns it into a dated commitment. Follow-up asks what happened and adapts the next move.
Start with a perspective
Mentors selected for function, not decoration.
19th century · leadership · persuasion · unity
Abraham Lincoln
President and writer whose speeches and letters reveal argument, coalition management, moral purpose, and restraint.
rich isolated corpus · source registration present Inspect mentor profile →18th century · economics · business · institutions
Adam Smith
Moral philosopher and economist on specialization, exchange, institutions, judgment, and commercial society.
rich isolated corpus · source registration present Inspect mentor profile →19th–20th century · business · wealth · organization
Andrew Carnegie
Industrialist and writer on enterprise, organization, wealth, and public responsibility.
moderate isolated corpus · source registration present Inspect mentor profile →4th century BCE · ethics · rhetoric · decision
Aristotle
Philosopher of ethics, rhetoric, politics, causation, and practical judgment.
rich isolated corpus · source registration present Inspect mentor profile →19th century · philosophy · psychology · motivation
Arthur Schopenhauer
Philosopher of will, desire, perception, character, suffering, and the disciplines of clear thinking.
rich isolated corpus · source registration present Inspect mentor profile →18th century · business · diplomacy · self-improvement
Benjamin Franklin
Printer, inventor, diplomat, and writer on self-management, enterprise, public cooperation, and persuasion.
rich isolated corpus · source registration present Inspect mentor profile →19th–20th century · education · enterprise · leadership
Booker T. Washington
Educator and institution builder on practical education, enterprise, leadership, and racial advancement.
rich isolated corpus · source registration present Inspect mentor profile →18th–19th century · strategy · conflict · leadership
Carl von Clausewitz
Military theorist on political purpose, friction, uncertainty, escalation, and strategic concentration.
rich isolated corpus · source registration present Inspect mentor profile →Or start with the objective
Clone a role-defined council.
leadership · judgment · communication
Become a powerful leader
Develop command, judgment, persuasion, resilience, and a vision others can understand.
5 seats · Command, Judgment, Persuasion, Resilience, Vision View the complete blueprint →persuasion · communication
Become genuinely persuasive
Build arguments people can follow while understanding the interests, dignity, and context of the audience.
5 seats · Rhetoric, Oratory, Human nature, Social influence, Framing View the complete blueprint →character · psychology · execution
Become the person your goal requires
Translate the objective into character, habits, capabilities, and a standard of daily conduct.
5 seats · Character, Habit, Self-command, Experimental craft, Moral courage View the complete blueprint →The trust contract
An advisory system should show its limits.
No corpus blending
A Marcus Aurelius answer does not quietly borrow from Machiavelli. Retrieval targets, citations, and release records remain mentor-scoped.
No invented authority
When retrieved evidence cannot support the question, the mentor must say so. A clean refusal is more useful than confident imitation.
Rights before release
Sources without a completed rights path stay out of production ingestion. Their public profiles remain clearly marked and leave the sitemap while blocked.
Private custom corpora
User-uploaded mentors live in private, user-scoped storage and indexes. Their files never become public pages or shared training material.