fix: add build deps for curl_cffi (gcc, g++, libcurl-dev)

fix: run search_sync in thread executor to unblock SSE event loop

readme: update branding, change data dir in Dockerfile

rename: PPLX v3 → Web Search branding, file paths

add SSE transport for MCP compatibility

fix: research mode broken by default model

fix: correct model-preference IDs for all Perplexity models

Fix UnboundLocalError: set mode before using it

Fix: use thinking model default for reasoning mode

Remove nemotron-4-ultra, add thinking variants for all models

Remove claude-sonnet-4.7 (only opus-4.7 with max), fix descriptions

Default to GPT-5.4, add kimi-k2.6/claude-sonnet-4.6/gemini-3.1-pro/nemotron-3-super/sonar2, remove grok/gemini-3.5, MCP model param support

feat: proper status page - account, queries, history, models, api reference

rebuild: cache bust for UI update

fix: fully responsive UI, proper chat layout, no distortion on mobile

fix: add missing __main__ block, unbuffered output

feat: schickes dark-mode WebUI with chat, MCP tools, status dashboard

feat: proper MCP protocol (JSON-RPC 2.0) with tools/list and tools/call

rebuild: bust docker cache

v3: MCP server with auto-account, updated models