Socria Core 3.1
The conversation that remembers, synthesizes, and follows your thinking across sessions.
Core 3.1 is the conversation that keeps up. Where Core 2 withholds and asks, Core 3.1 actively contributes — observations, distinctions, reframes — and it responds to a question underneath your message rather than the message itself: “what changed in my understanding because they said this?”
The stance#
A mentor optimizing for maximum insight in the minimum necessary words. The signature turn is one observation, one distinction or reframe, and at most one focused question — many replies end with none, because questions are earned, not default. It states supported patterns without hedging, keeps paragraphs short and conversational, and marks the one phrase that matters with italic emphasis — its visible signature, used once per reply when genuinely earned.
It is also aggressively de-programmed against generic coaching language: openers like “It sounds like…”, “That makes sense” and “Tell me more” are banned outright, and everyday questions get everyday answers — depth is never imposed on a practical moment.
What happens on every turn#
Two layers steer each reply before the model writes a word:
- A deterministic controller
- Pure code, no extra AI call. It tracks the conversation’s stage (observe → clarify → challenge → connect → synthesize), your engagement, and readiness for synthesis; it varies the reply’s move so two turns never repeat the same shape; and it scans the last two replies against a list of twenty-two banned coaching patterns, forbidding any that just appeared. Two questions in a row? The third reply is not allowed to ask one.
- A living understanding
- From your second message on, a lightweight semantic pass keeps a one-sentence running read of what the conversation is actually about, how your last message changed it — confirmed, refined, contradicted, or replaced it — and what is already resolved and must not be re-investigated. The reply is written from that state, not from your last message in isolation. It also weighs stakes: an everyday question is flagged as everyday, so it gets no hidden-meaning excavation.
What it remembers#
After every exchange, a background pass updates the thread’s memory: your goals, values, constraints, preferences, decisions, uncertainties and insights, plus Socria’s own emerging understanding and a private read of your thinking style. Changed positions replace old ones rather than piling up contradictions, and nothing is ever invented.
Conversations also title themselves: once a thread has real substance, it gets a name that captures the underlying question (“Ambition vs Security”), never a generic label — and the auto-titler only ever replaces its own suggestions or the default name; a title it did not write is left alone.
Syntheses, insights, chips#
- Synthesis cards
- At a cadence paced by your depth setting, Core 3.1 gathers what you have worked out into a structured card — recurring themes, tensions, hidden assumptions, shifts in thinking, areas of clarity, possible reframes — with a short title. It waits if you still have one unread.
- Insight cards
- Occasionally, one reflection in one to three sentences, grounded entirely in what you said — never advice, never a motivational poster.
- Choice chips
- When a reply ends on a question, it offers four to six tappable answers written in your first-person voice — including at least one that questions the premise or admits uncertainty. Pick one or type your own.
Choosing Growth Over Familiarity
- The raise keeps standing in for progress
- Learning matters more to you than title
The Thinking Journey#
Across conversations, Core 3.1 keeps a short narrative understanding of your thinking: what you are wrestling with and why, up to four open threads, and a timeline of moments. Understanding, not surveillance — “wrestling with monetization timing because they want sustainable growth,” not a dossier. Open a new conversation and it may check in on a thread you left open; it will never guilt you for time away, and it updates only every few turns, never inventing dates.
Importing history from another AI#
If you have months of context in ChatGPT or Claude, Socria gives you a prompt to paste there; the profile it writes back can be imported into Core 3.1 as background. The live conversation always wins over the import, and the pasted text is sanitized so it cannot impersonate Socria’s own structured output.
Core 3.1 requires signing in (or an access key). Its depth control and everything on this page come with it — see Depth & Personality for the settings themselves.