How sessions work
A SynthBoard session is your standing board of expert AI advisors, debating one decision in real time. Here is what each of the concepts means and how they fit together.
Two rooms#
Every session runs in one of two rooms.
The Boardroom
A staffed board of advisors. An adaptive intake turns your question into a decision brief, the brief staffs the board, the board researches and debates, and the run ends in a decision memo or a shaped deliverable.
The Consult
A 1-on-1 with a single named expert. Conversational and billed per turn — the expert answers, pushes back, and composes real documents when the ask calls for one. Escalate to the full board when the stakes justify it.
Synths#
A synth is a specialised AI advisor with its own role, personality, and cross-session memory.
Pick 2–12 synths per session from a library of 24 curated archetypes — CFO, Engineer, Skeptic, Customer, Devil’s Advocate, Strategist and so on — plus any custom synths you have created. Each synth has its own role, its own personality (risk tolerance, push-back level, language register), and its own memory across your sessions.
Each synth contributes their own response per round, drawing on a frontier reasoning model picked to fit their personality and the mode. Override the model on a synth-by-synth basis when you want tighter control.
Modes#
The 10 branded modes shape how the board engages your question and how the synthesis converges.
The Boardroom
Collaborative advice from complementary experts.
Idea Lab
Divergent ideation without early critique.
Lightning Round
Fast, sharp opinions in under 100 words each.
The Arena
Adversarial exploration of opposing positions.
Shark Tank
Ruthless attack on an idea to surface every flaw.
Command Center
Structured framework for choosing between options.
War Room
Crisis triage with immediate action plans.
Deep Dive
Thorough expert analysis with evidence.
Crystal Ball
Scenario planning with probability-weighted outcomes.
Research Lab
Evidence-first synthesis across disciplines.
The board flow#
From question to memo — the phases every board run moves through.
- Intake. A facilitator asks only what it actually needs and drafts a decision brief — the framed question, the known facts, and the deliverable shape.
- Staffing. The brief seats the right advisors for the question. You see who is on the board, why, and the credit estimate before anything runs.
- Research. Each seated advisor investigates independently — web search, your uploaded context, their own knowledge base.
- Debate. The board argues genuinely opposed positions; disagreements are surfaced, not averaged away.
- The memo. Decision boards deliver a decision memo — verdict, confidence, the conditions that would flip it, and the strongest dissent. Other asks deliver an artifact shaped to the question: a plan, a forecast, a comparison, a research brief.
Credits#
Usage-based pricing — you only pay for what runs.
Each synth response is metered against the model it ran on, plus the synthesis pass at the end. If you enabled web search or tool execution, those add their own usage. The full breakdown lands on the session page once it completes.
For tier comparisons and a per-mode cost feel, see Pricing & credits.
Tools#
Synths can read from your integrations and propose actions.
With a connected integration, your synths can pull context (a Gmail thread, a Linear ticket, a HubSpot deal) and propose actions (drafting a reply, creating a ticket, updating a record). Every action flows through the same governance pipeline:
- The synth proposes the action in their response.
- The Skeptic reviews it for risk before it ever shows.
- You see the proposed action with a clean preview.
- You approve → it executes → you have a 30-second undo window.
Output#
Every session produces a structured, shareable record.
- The decision memo (or shaped artifact) — the board’s terminal deliverable.
- Each synth’s positions and evolving claims.
- Cross-round summary tracking how positions shifted.
- Optional action plan with structured next steps.
- Optional decision record for when you formally commit.
- Bookmarks — anything you pinned during the debate.
- Exports as Markdown, JSON, or plain text.
Transports#
One engine, many ways in.
Sessions run from the web UI, from any MCP client (Claude Desktop, Cursor, Windsurf, Zed, ChatGPT), from the REST API, or from automation platforms (Zapier, n8n, Make, Pipedream, Activepieces, Arcade) via the API and webhooks — native adapters for those platforms are in development. Every transport hits the same engine, persists to the same record, and shares the same credit balance.