Soqratic. A dialogue API that asks the next right question.
Most chat APIs reward fluency. Soqratic rewards inquiry — anchored to the documents on the table, paced by a posture, audited on every turn. Long-horizon memory, scaffolded questioning, and a structured rationale attached to each question the system proposes. Used by companies running review meetings, by teams in design critique, by advisors interviewing stakeholders, and by individuals working through their own thinking. Less chatbot, more a colleague who reads carefully and asks well.
The opposite of a chatbot.
Soqratic is a conversation API where the system’s job is not to answer, but to ask. Each question is generated in light of the conversation, the documents on the table, and the intent of the session — and is returned with a one-line rationale you can read at a glance. Run it as a self-paced conversation, run it with a host who edits each question before it ships, run it on autopilot. The same primitive serves all three.
You configure a session with a document or a small library, an intent (review, interview, debrief, decision), and a posture (curious, sceptical, time-bounded). Each turn returns the next question, the rationale behind it, and any flags — places where the conversation is drifting, where the documents and the participant disagree, or where a decision has been deferred too many turns in a row.
Three rooms it works in
- Inside a company. Board debriefs, post-mortems, audit reviews, strategy reviews. Soqratic asks the questions that get a room past polite agreement, surfaces what the document does and does not actually say, and pushes from reaction to decision.
- Across a team. Design critiques, hiring conversations, customer interviews, research debriefs. A host runs the session — often a facilitator, a researcher, a hiring lead — and Soqratic keeps the line of questioning sharp, holds onto threads people try to skip, and flags where what is said does not match what was written.
- For one person. Working through your own draft, your own report, your own decision. Soqratic interviews you about the thing in front of you, with the same depth a good colleague would, and lets you hear yourself think.
Per-conversation, not per-token.
Soqratic is sized by the session, because that is the unit that matters. A session is a single conversation up to 90 minutes or 80 turns, whichever comes first. Long context is included. Memory across sessions, scoped to a participant and a document set, is included. Pricing is part of the enterprise contract; volume curves are posted on request.
- Sized for small teams & pilots
- Configurable session-memory retention
- Default rubric library
- Observatory traces included
- Higher volumes, dedicated capacity
- Long-horizon session-memory retention
- Custom rubrics, postures & question pools
- Webhooks, exports, enterprise governance
- Private deployment, your region
- Retention shaped to your policy
- Enterprise governance options on request
- Onboarding workshop included
Three primitives.
Open a session against the documents you care about, take turns, read the transcript with the rationale attached. Each proposed question carries a one-line reason. Run it self-paced, or attach a host (a facilitator, a researcher, a manager) who reviews the question before it ships.
POST /v1/soqratic/sessions
{
"documents": ["report:audit-q1"],
"intent": "review", // review | interview | debrief | decision
"posture": "curious-sceptical",
"host": "host-22", // optional — omit for self-paced
"participants": ["user:ceo"]
}
POST /v1/soqratic/sessions/:id/turn
{ "participant": "We feel the recommendation is solid." }
// → returns the next proposed question, its rationale,
// any flags, and an editable draft you can ship as-is.
GET /v1/soqratic/sessions/:id
// → returns the full transcript with rationale on every turn.Named use cases that ride the same primitive.
The questioning engine is general; the workflows people want are not. We ship a small number of named presets — tuned postures, default rubrics, surfacing rules — and customers add their own in week two of an install.
- Journaling. A daily or weekly reflection cadence with the “dive deeper” follow-up pattern — the next question lands not in a chat box but as a deepening prompt against the entry the participant just wrote. Used by us internally for facilitator pulse-tracking; portable to any reflective-practice context.
- NorthStar. A mission-and-vision dialogue preset. Long-horizon memory, signals as context, versioned drafts. The participant gets coached toward an articulation, not handed one.
- Review & debrief. Post-mortem and decision-review postures with “what is this document avoiding” and “what has been deferred” flags surfaced on every turn.
- Custom. Most installs ship one or two domain-specific presets we write together in week two — interview, intake, hiring read, board pre-brief. The preset is code; the harness is the same.
Stop building inquiry bots from scratch.
If you are wiring a chat loop on top of a document and writing system prompts about Socratic method, you are about to spend six months on what we already have.