BrainForge/Security: direct vLLM (4090), HANA persona cache, Security stub
NeonClary
23ed0c3
- Config: neon_security_vllm_base_url, HANA_KLATCHAT_PASSWORD aliases, supplement ids
- hana_client: merge vLLM /v1/models, enrich Security personas from HANA, stub if missing
- llm_router: prepend HANA persona base for neon_direct_vLLM Security path
- Orchestrator/chat: pass vLLm fields through resolve_model; lifespan warms persona cache
- .env.example: document vLLM key vs HANA login
Made-with: Cursor
Add HANA klatchat credentials for BrainForge Security (4090 x1-3)
NeonClary
794d51e
Add Mistral.ai as direct provider with Mistral Small 3.2, Small 4, and Devstral2
NeonClary
4d765a8
Made-with: Cursor
Simplify conversation ending: two clean closing messages after winding_down
NeonClary
49ac02e
Made-with: Cursor
Strip thinking content from all LLM responses and skip redundant role generation
NeonClary
6d068d4
Made-with: Cursor
Add AI-completed vs exact role generation style toggle
NeonClary
4f7950f
Made-with: Cursor
Improve freeform role generation prompt and expose finish_reason
NeonClary
4fd9d4c
Made-with: Cursor
Add response timing, chat stats, and role prompt viewer to settings
NeonClary
03151b9
Made-with: Cursor
Rebrand UI with Neon.ai header, sidebar styling, expert personas, and UX improvements
NeonClary
0f3e9c6
Made-with: Cursor
Add speed-priority setting with 5s racing fallback for faster responses
NeonClary
2845211
Made-with: Cursor
Add Neon HANA routing, per-request timeouts, and chat brevity prompts
NeonClary
b49a6ce
- Add llm_router.py to unify LLM calls across OpenAI-compat and HANA backends
- Cache HANA persona system_prompts on startup to fix 422 errors
- Add per-request timeout support: 20s chat, 45s role gen, 120s floor for thinking models
- Add 4x max_tokens multiplier for thinking/reasoner models to prevent truncation
- Add brevity instruction to all conversation prompts (2-4 sentences, chat-style)
- Add anti-reasoning directive to suppress chain-of-thought leaking into responses
- Reduce default chat max_tokens from 1024 to 500 for faster responses
Made-with: Cursor
Add HuggingFace Spaces deployment with OAuth and rate limiting
NeonClary
c555047
Made-with: Cursor
Add freeform persona input mode with file upload and Developer menu toggle
NeonClary
1a97647
Made-with: Cursor
Add Developer menu with orchestrator picker, download buttons, and LLMChats2-style dropdown
NeonClary
3e23f7a
Made-with: Cursor
Initial commit: LLMChats3 - LLM conversation app with React + FastAPI, persona config, orchestrated conversation lifecycle, collapsible sidebar accordions, stop button, and export capabilities
