fix: raw passthrough proxy — preserve tools, tool_choice, and all OpenAI params

feat: live model list from upstream API instead of stale DB cache

fix: update NVIDIA endpoint key on startup if env var changed

refactor: pure model proxy - remove RAG/web search from bridge endpoints

Fix streaming: parse stream_llm SSE deltas before wrapping in OpenAI format

Fix: use build_chat_url for correct URL construction in OpenAI bridge

Register OpenAI-compat bridge router + exempt /v1 from session auth

Add OpenAI-compatible bridge endpoint for Hermes integration

fix: set search_url for SearXNG + add hard_max

chore: increase agent_input_token_budget to 100000

fix: remove set -e from entrypoint, fix function ordering in auto_configure

auto_configure: enable Playwright browser MCP on startup

Add auto-configure + improved Telegram bot bridge

Add Telegram bot bridge + optimize agent settings

Fix: use --no-build-isolation for SearXNG install

Fix: pre-install msgspec for SearXNG build

Deploy Odysseus AI Workspace with ChromaDB + SearXNG + ntfy

initial commit