Skip to main content

Round Table

Round Table runs a multi-persona AI deliberation across a topic. Pick personas (Moderator, Devil's Advocate, Senior Developer, QA Lead, Product Owner, Designer, or your own), send a message, and each persona replies in turn before they deliberate with one another. Use it to debate decisions, stress-test plans, or run a "design review" without the meeting.

When to use a Round Table

  • You're stuck between two designs and want a structured debate
  • You want adversarial review (Devil's Advocate) before committing
  • You need multiple specialist perspectives (QA, PM, designer) on a single artifact
  • You're prototyping a tricky feature and want a pre-implementation gut-check

Configuration

When creating a round table:

SettingDescription
WorkspaceRepo context for the discussion
BranchOptional branch checkout for the worktree
AI ProviderWhich provider runs the personas
AI ModelSpecific model
PersonasAt least 1, max 10
Owner-PrivateOptional — only the creator sees this round table

Personas

Pre-defined personas

PersonaRole
ModeratorKeeps the discussion on track
Devil's AdvocateChallenges assumptions
Senior DeveloperArchitecture, maintainability, technical risk
QA LeadEdge cases, test strategy
Product OwnerUser value, scope, priorities
DesignerUX, IA, accessibility

Custom personas

Create custom personas at the workspace level (e.g., "Security Auditor", "DevOps Lead"). Each persona can be enabled/disabled per workspace and is soft-deleted when removed (custom personas only).

Personas are snapshotted on first message — once a discussion starts, edits to the persona definition don't retroactively change the round table.

Manage Personas page

Use the dedicated Manage Personas page to create, edit, enable/disable, and soft-delete custom personas across all workspaces you have access to. The same page is where you toggle pre-defined personas on or off per workspace. Access is gated by the roundtables.manage_personas permission.

Turn flow

When you send a message, each enabled persona weighs in, the personas may then deliberate with one another, and the round table surfaces a few suggested next messages — phrased from your perspective — for you to continue with.

Concurrent sends to the same round table are blocked while a turn is in flight, so each turn completes before the next begins.

Turn trigger kinds

KindEffect
Send MessageStandard user message → persona replies
RetryRe-runs a failed turn
VoteCalls a vote (see below)
SummaryGenerates a fresh running summary

Each turn has a status: Processing, Completed, or Failed.

Call a Vote

Trigger a structured vote among the personas:

  • Question (1–500 chars)
  • 2–6 unique options, or Yes/No preset
  • Each persona casts one vote
  • The tally is persisted as a vote-result message in the discussion

Use votes to force a decision when the discussion is going in circles.

Running summary

Generate a fresh summary at any point — Polygent runs a dedicated turn that produces a concise digest. The latest summary is highlighted, and a marker is added to the message stream so you can see when it was created.

Reactions

Add Pending reactions (Agree / Decline) to your own messages. They are bound to the user message at finalization on the next turn. Pending reactions can be removed before the next turn.

Drafts (no persistence)

While reading a round table, draft a ticket or plan from the discussion:

  • Ticket Draft — opens the ticket creation modal, pre-filled
  • Plan Draft — opens the plan creation modal, pre-filled

Drafts are not persisted — close the modal without saving and the draft is gone.

Linked artifacts

Tickets and plans created from a round table back-reference the discussion they came from, so you can trace every ticket or plan that originated from a round table.

Export

Export the full discussion as:

  • TXT — plain text transcript
  • Markdown — formatted for docs

Lifecycle

StatusMeaning
ActiveOpen and accepting turns
CompletedClosed
HiddenDismissed from the sidebar

Statistics

A statistics panel shows per-round-table:

  • Turns by status / by trigger
  • Daily trend
  • Top models
  • Top users

Plus an audit trail with full turn history.

Worktree management

Each round table gets its own worktree, verified on startup and cleaned up if orphaned. Owner-Private round tables are inaccessible to other users.

Live updates

All connected viewers see real-time updates — new persona replies, vote tallies, summary generation, and status changes appear automatically without a page refresh.

Examples

Stress-test an architecture decision You're choosing between a message queue and direct API calls. Start a round table with Senior Developer, Devil's Advocate, and QA Lead, and post the two options. The Senior Developer weighs maintainability, the Devil's Advocate attacks each assumption, and the QA Lead raises failure modes. Call a Vote with the two options to force a decision, then create a Plan Draft from the discussion.

Pre-implementation design review Before building a new screen, run a round table with Designer, Product Owner, and Senior Developer against a mockup. You get UX, scope, and feasibility feedback in one pass — then convert the agreed direction into a ticket with one click.

Resolve a stuck debate A thread keeps going in circles. Generate a Running Summary to capture where things stand, then Call a Vote (Yes/No preset) to settle it. The tally is recorded in the discussion as a permanent decision record.

Add a domain specialist Create a custom Security Auditor persona at the workspace level, enable it, and include it in a round table reviewing an authentication change — its perspective is snapshotted into the discussion so later edits to the persona don't rewrite history.

Permissions

PermissionCapability
roundtables.createCreate new round tables
roundtables.manage_personasAdd/edit/disable workspace personas