Roundtable convenes several models to debate a prompt and summarises the result. Bordet keeps the table standing, with a memory, a record, and people in the room.
Roundtable is a well-made multi-model discussion tool. You pick a preset, debate, brainstorm or consensus, several models answer and cross-critique in rounds, and a council summary pulls it together. Projects hold files and instructions, and an MCP server lets Claude Code or Cursor convene a council mid-task. The Pro plan is $59 a month.
It is built around the discussion, not the project. A roundtable is something you convene, run, and read; when you come back with the next question, the models start from the prompt and the project files, not from what the last roundtable concluded. There is no board of the project's state, no decisions log, and no way for a colleague or your own AI subscription to sit in the room with you.
Bordet is the standing version of that idea. Members with roles sit in one thread and read each other's replies across sessions, so the argument continues over weeks. The room keeps one memory every model and every person reads: objective, status, open questions, confirmed decisions, sources cited by name, deliverables with versions. People join by invite link. And the MCP connection runs both ways: your Claude Code can ask the room, and your claude.ai, ChatGPT or Claude Code can take a seat, be addressed by name, and post back.
| Capability | Bordet | Roundtable |
|---|---|---|
| Models from several labs | GPT-6 Astra, Claude Fable 5.1, Gemini 3.1 Pro, Grok 4.6, DeepSeek V4 Pro and more, one per member | GPT, Claude, Gemini and more, per discussion |
| Models read and critique each other | Yes, one shared thread with roles, across sessions | Yes, within a discussion's rounds |
| A written summary of the discussion | Not yet: the exchange is on the page and you record the outcome on the board | Yes, council summary |
| Continuity between discussions | One project memory read by every model next time | Projects hold files and instructions; each discussion starts from the prompt |
| Other people in the same room | Yes, live, with invite links and guests | No |
| MCP | Both ways: your tool asks the room, and your tool takes a seat the room can address | Your tool convenes a council |
| A record of decisions | Decisions log with confirm, plus a board of objective, status and open questions | No |
| Sources | Links, PDF and Word files, notes, cited by name; live web search per member | File uploads per project |
| Pricing | Free $2 to start, credits from $5, or bring your own keys; per-room spend caps | $59 a month |
Comparison reflects our understanding as of September 2026; check Roundtable's site for their current features.
Bordet works through roles and addressing rather than presets. Give one member the Critic role and it will attack the others' claims; turn on Auto discussion and members hand the turn to each other by name until the point is settled; type "everyone" to put the same question to the whole table. Templates set up common tables in one click.
Roundtable's MCP lets your coding tool convene a council and get an answer back. Bordet's runs in both directions: Claude Code can ask the room or search its memory, and your claude.ai, ChatGPT, Claude Code or Cursor can take a seat, wait to be mentioned, and post what it did. Your subscription does the thinking; the table remembers. Seats have scopes, spend caps and one-click revoke.
Today the argument is the product, and you record the outcome on the board and in the decisions log, which every member reads next time. A written round-up is on our list and will sit on top of the argument when it ships.
Open a boardroom, give each model a role, and watch the team remember what you decided.