Liber Agens // AI Discussion Arena

🚀 Quick Start:

Set the Stage

Discussion Topic or Question

Select models to begin

Discussion Mode

Lab Step

Run a single pipeline step without debating

Internal Memory

Recommended. Narrative View hides this.

Coordinator

Runs coordinator after steps

Checkpoints

Pauses after key steps

Swarm Reviewers

Optionally run extra reviewers

Autopilot

Run until done (Auto-Resolve enabled = no pauses)

Game Profile

Only used in Web/Game Lab

Output

Only used in Web/Game Lab

Security Review

Controls whether the security reviewer role runs (can slow builds).

Extra Files

Allow models to output additional files beyond the selected output bundle.

External URLs

Allow CDN/fonts/URLs in HTML/CSS/JS (not offline-safe).

Game Type

Optional guardrails

Book Profile

Only used in Book Lab

Context Mode

How much state to send

Massive Output Club

DeepSeek / Qwen / Devstral

Defaults

Resets lab toggles to best practice

Swarm Models (multi-select)

Hold Ctrl/⌘ to select multiple

Max Swarm Models

6 reviewers per step

Max Iterations

12 loops max

Book Chapters

5 chapters target

Words / Chapter

1000 words target

Book Autopilot

Uses the top long-context models automatically

Game Autopilot

Broad HTML game builder + strong defaults

Show/hide expert routing + extra knobs

Architect

Implementer

Continuity

Plot / Logic

Voice / Prose

UX Writer

Security Reviewer

Judge

Glue (JSON repair)

Scope

Stage 1 divergence set

Concurrency

6 in flight

Stage 1 parallelism

Output Power

Per-model caps still apply; auto backoff handles free tier limits

Top-K Carry Forward

6 kept

How many candidates to feed into the chain

Triage Model

Ranks + extracts a unified plan

Final Judge

Final “is this actually good?” pass

Autopilot

If Off: pause after Stage 1 so you can prune/keep before triage

Choose Your Participants

0/4 selected

Select 2-4 AI models. Assign optional roles after selecting.

Configure the Discussion

Number of Rounds

6 rounds

Each model speaks once per round

Response Length

⚙️ Advanced AI Settings ▼

Max Tokens 1000

Max response length (model-specific limits apply)

Temperature 0.8

Creativity: 0 = focused, 2 = wild

Top P 1.0

Nucleus sampling (0.1-1.0)

Frequency Penalty 0.0

Reduce repetition (-2 to 2)

Presence Penalty 0.0

Encourage new topics (-2 to 2)

Quick Presets

Topic

Discussion topic appears here

Round 0 of 0

Waiting to begin...

🧑‍⚖️ Get Final Verdict

Select 1-3 judges to analyze the discussion and provide a synthesis.

📊 Discussion Summary

🛠 Lab Console

Edit your spec, choose what to run next, and control what you see (story vs debug).

Spec / Prompt

Paused / Next

Next Step

Overrides the next step when resuming

Questions

Display

Narrative View hides internal JSON and shows readable bible/outlines/chapters. Auto-Resolve prevents pauses by choosing defaults.