N
New Conversation
0
Loading…
…
Compare
0
0
Compare Arena
A:
Primary
vs B:
Grounded on shared citations
Suggested follow-ups
Send
Enter
to send ·
⇧Enter
newline
All responses grounded in your local ERCOT corpus — no data leaves the device.
Settings
Provider, model profile, and retrieval tuning
Inference Provider
MiniMax (Cloud)
OpenCode Go (Cloud / Unified)
Bifrost Gateway (Local)
Local llama-server / vLLM
Custom OpenAI-compatible
Model Profile
Refresh
LightRAG Backend Base URL
LLM Synthesis Endpoint
API Key / Token
Retrieval Mode
Hybrid Mix (KG + Vector)
Dense Hybrid
Vector Only
Chunk Top-K
Max Completion Tokens
Managed Inference Providers
0
Used by Compare Arena to pick models across providers
Add Provider
Seed Defaults (MiniMax, OpenCode, Bifrost)
Compare Arena — secondary model (optional)
Default Secondary / Compare Model
Compare Endpoint (optional)
Compare API Key (optional)
Auto-tuning:
selecting a model profile automatically calibrates Top-K and completion token ceiling using empirical leaderboard limits.
Cancel
Save & Apply
ERCOT Nodal Protocols
71
ingested documents · search and read any section
Close
Document Viewer
Loading from local corpus cache…
Rendered
Raw
Open .docx
.docx
.md
.txt
Copy
0/0
Loading…
Map
Research Dossier
Pinned regulatory excerpts · search, export as Markdown
Export .md
Close
Prompt Templates
Click to insert into the prompt bar
Close
Create Collection
Group related regulatory sessions into a dedicated workspace
Collection Name
Description (optional)
Folder Color Accent
Cancel
Save Collection
Add to Collections
Select collections for this conversation
New Collection
Done