fix(proxy): architectural overhaul for timeouts, 9k limits, and infinite hang prevention

fix: change rate limit cooldown from 5 minutes to UTC midnight

refactor: remove manual XML parser to support native Cloudflare tools and implement sequential account rotation

feat: set kimi-k2.6 as default model in mock models response

fix: mock /v1/models response to bypass Cloudflare 405 Method Not Allowed error

fix: resolve stream dropping issues (fix hit_limit logic, flush incomplete XML buffers, add missing Anthropic Kimi reminder)

fix: flatten array content to string in OpenAI endpoint for Cloudflare Zod compatibility

fix: resolve OpenAI invalid format for API errors and show decimal neurons in UI

fix: resolve absolute path to dashboard.html

fix: remove global asyncio.run() to avoid RuntimeError in uvicorn

feat: add professional dashboard UI and implement Cloudflare Account Neuron limits

feat: implement XML tool call parser for Kimi K2.6

feat: inject strict system reminder to force Kimi model to use XML tool tags before generating

chore: add logging for final generated text length on successful stream completion to aid debugging

fix: ensure proxy does not strip chunks containing tool_calls, restoring function calling capabilities

fix: maintain consistent stream_id and created_timestamp across auto-resumes in OpenAI endpoint to prevent client stream interruption

fix: mask finish_reason in OpenAI stream format to prevent client from prematurely closing connection during auto-resume

feat: make Kimi stream auto-resume completely seamless by handling CF length truncation and accelerating key search

fix: resolve unmatched parenthesis syntax error at the end of proxy_cf.py

fix: prevent Kimi model truncation by increasing max_tokens and adding stream auto-resume in proxy_cf.py

Update proxy_cf.py

Upload proxy_cf.py

Delete proxy_cf.py

Update proxy_cf.py

Upload proxy_cf.py