Improve local/self-hosted model deployment experience

[wyuc]

Several issues report friction when using locally deployed models (Ollama, private LLM/TTS/ASR endpoints). Collecting them here for tracking.

Related issues

• 使用本地部署的模型后会出现拒绝访问 #310 SSRF guard blocks private network URLs
• [Feature]: 对使用私有模型部署的，需跳过安全机制可在readme中添加简要说明 #311 Request for docs on bypassing SSRF for self-hosted
• Bug: Using Ollama OpenAI-compatible endpoint causes "LLM returned empty response" #271 Ollama OpenAI-compatible endpoint returns empty response
• [Feature]: ASR和TTS支持openAI 兼容协议 #292 TTS/ASR lack custom OpenAI-compatible provider support
• Discussion Local LLM or Local TTS #106 Local LLM / Local TTS questions

Root causes

1. SSRF guard blocks all private addresses. Necessary for SaaS, but blocks self-hosted users. LLM routes skip the check in dev mode, but TTS/ASR/image/video routes check in all environments.
2. No first-class local provider support. LLM has custom provider mechanism but Ollama compatibility is unverified. TTS/ASR have no custom provider mechanism at all.

Planned work

• P0: Trust server-configured URLs in SSRF guard (addresses configured via .env.local or server-providers.yml bypass SSRF check), or add ALLOW_LOCAL_NETWORKS env var. Fixes 使用本地部署的模型后会出现拒绝访问 #310 [Feature]: 对使用私有模型部署的，需跳过安全机制可在readme中添加简要说明 #311
• P1: Review and merge PR feat: Ollama local LLM support #94 (built-in Ollama provider). Fixes Bug: Using Ollama OpenAI-compatible endpoint causes "LLM returned empty response" #271
• P2: Add custom OpenAI-compatible provider for TTS/ASR. Fixes [Feature]: ASR和TTS支持openAI 兼容协议 #292