RoundTable
Council of Minds · Private beta

This build is closed for testing. Enter the access code to continue.

Incorrect code.
⚠ Deterrent only — not encryption. A determined visitor can bypass this.

Welcome to the Round Table

A council of minds, convened.

⚠

Before you begin

Set usage caps on your API accounts. RoundTable sends your prompts to whichever models you activate, on every turn. Costs can add up quickly across multiple models and rounds. Set hard spending limits in each provider's billing console — RoundTable is not responsible for any overuse of credits or unexpected charges.

Your keys stay in your browser. If you enable “Remember on this device,” API keys are stored in this browser's local storage and never sent anywhere except directly to the providers you call. Only enable that on a private device.

This is a direct-key build. By default your keys talk straight to each provider from your browser. You are responsible for keeping them safe — clear saved secrets from the sidebar when using a shared machine.

RoundTable
Council of Minds · v1
Participants Gemini · image studio 0 active
Session
Mode
Suggested rounds
Turn order
Advanced: use proxy Optional. Leave off unless you've been given a proxy URL and bearer token. Direct API keys above are the normal flow.
Blind first round Each model writes their opening turn without seeing the others — kills first-mover bias.
Send full code & files to bots By default, uploaded files and zips are truncated (≈80 KB per file, 500 KB total) before being sent to the models — this caps token cost. Turn this ON to send the FULL content of every attachment. ⚠ This can dramatically increase token usage and cost on large files or zips — a single big repo can run to hundreds of thousands of tokens per bot, per turn.
Recall past sessions Let the council read transcripts from earlier debates/collabs you select, so they can build on prior work. Choose which sessions with the button below. Adds their text to every bot's context — more sessions means more tokens per turn.
Permissive mode (Gemini) Sends safetySettings: BLOCK_NONE to Gemini text + image calls. Significantly loosens text refusals. Image refusals only reduced at the edges — Google's image filter is partly hardcoded. No effect on Claude, GPT, or Grok (their APIs don't expose this). Provider-side policy and model-level refusals can still apply. Operator owns what is generated.
Gemini Image Mode Swaps the workspace into an AI Studio-style Image Studio: prompt → image directly (no council, no turns). Pick model, aspect ratio, resolution, and 1–4 images per run. Edit prompts, delete outputs, use outputs as references for iterative editing. Uses your Gemini key. Toggle off to return to the round table.
Remember on this device Saves keys & setup to browser storage. Only enable on a private device.
The Round Table
Image Studio
Awaiting topic
Direct image generation · Gemini
0 tok · $0 Round 01

Convene a council of minds.

Pose a question, upload evidence, and let the activated models deliberate. Each takes a turn. You hold the gavel.

Topic
Viewing archived session · read-only · participant controls disabled
The chamber is silent.
Activate at least two participants in the sidebar, add their API keys, then pose a topic above to begin.
Image Studio uses Gemini image models only in this build.
References
No references — generations will be prompt-only. Add images of your subject for identity-consistent results.
Image Studio
Pick a model, ratio, and resolution above. Add references if you want identity consistency. Type a prompt below and Generate. Uses the Gemini API key from the sidebar.
Idle
Error & Issue Log
Problems logged this session — newest first. Not saved between reloads.