Private Local Lab · v0.6.16

AIm³ MentalArena Lab

Local council console for BD’s multi‑LLM RHP/RHPr/RHPm workflow. Default mode is now Top‑5 Council: ChatGPT, Claude, DeepSeek, Gemini, Grok. Single-provider calls are only a debug/smoke-test tool.

not checked
no run

Task seed

Define the problem once. MentalArena expands it into the working prompt.

Title BD seed / problem

Protocol profile

Output type

Start here. After a run starts, the expanded prompt appears in Step 2.

Default Council: Top 5 LLMs

This is the main path. It sends the same prompt to GPT, Claude, DeepSeek, Gemini and Grok.

loading provider status…

Provider status not checked yet.

Council target group

Active stage (auto)

You no longer need to change Batch stage manually. Buttons below switch it automatically.

Current send target: Round 1

stage guard on

Round prompt / editable prompt

Advanced: manual stage override

Only use this for recovery/debug. Main workflow changes stage automatically.

Manual stage

Batch results will appear here after a council send.

Single-provider debug / smoke test

Use this only to test one API key/model at a time. It is not the main MentalArena workflow.

Provider

Model

Endpoint

Allow manual model/endpoint override for this debug call. Leave off unless you intentionally want a custom model. Direct prompt

Results, conflict map, critique, judge, synthesis

After Round 1 answers exist, build the next MentalArena layers.

Ready.

Output

Use Step 1 to start a run. Then use Step 2 to send the prompt to the default Top‑5 Council.

Artifacts

No artifacts yet.

Direct result no call

Only for single-provider debug. Smoke tests verify the API route with a nonce. Provider/model identity comes from API metadata below, not from the model’s self-report. Provider-API.txt is secret-only; Provider-Config.txt stores model/endpoint/settings. API key files are never overwritten by patches. Use Copy result; no manual selection needed. Route-smoke verifies the API path by echo token; model identity comes from API metadata, not self-report.

No direct call yet.

Direct call metadata / log files

No direct call metadata yet.

Manual import

Provider/name

Paste answer

What this lab does

This lab automates the manual workflow: seed → prompt expansion → LLM answers → conflict extraction → targeted critique → judge prompt → final synthesis prompt. Python keeps the ledger; LLMs do semantic judgement; BD keeps final override.

Main change in v0.6.16: guided workflow stage control was added. The Lab now auto-selects Round 1 / Round 2 / Judge / Final stages, prevents silent overwrites, and shows a floating checklist. Route-smoke remains available under debug.

All provider status

loading…

Security

Do not put real API key files into GitHub/Netlify. For production later, use Netlify Functions + environment secrets. This local lab is for localhost use.