feat: support OpenClaw provider ecosystem (minimax-portal, ollama, custom) in memos-local-openclaw

[Ink-kai]

Summary

memos-local-openclaw fails with multiple provider combinations common in OpenClaw deployments:

1. minimax-portal: OpenClaw default, uses Anthropic API, but switch cases only have explicit "anthropic"
2. Ollama /v1/embeddings: Broken OpenAI compat endpoint, fallback loses vector embeddings
3. Ollama /v1/chat/completions: Broken OpenAI compat endpoint, summarizer timeouts
4. Timeout too short: judgeDedup default 15s insufficient for local model loading

Root Causes

1. Provider switch cases only have generic names ("anthropic"), missing custom provider keys like "minimax-portal"
2. Ollama OpenAI compatibility broken for both /v1/embeddings and /v1/chat/completions (native APIs work)
3. Default 15s timeout too short (Ollama loads 2-6s per cold request)

Proposed Solutions

1. Add explicit cases for common OpenClaw providers

Add minimax-portal and other known providers to all 6 switch functions:

case "anthropic":
case "minimax-portal":
  return summarizeAnthropic(...);

2. Support Ollama native API

case "ollama":
  return embedOllamaNative(texts, cfg, log);  // uses /api/embeddings
  return summarizeOllamaNative(text, cfg, log);  // uses /api/generate

3. Increase default timeouts for local models

• judgeDedup: 15s → 120s
• summarize: 60s → 120s

4. Add startup health-check

Fail fast if configured providers are unreachable.

Related

• Feature request: add native Ollama provider support for memos-local-openclaw #1231: Ollama native support
• feat: add MiniMax as a first-class LLM provider #1250: MiniMax first-class provider
• fix: summarize failed, timeout error: the operation was aborterd due to timeout #1194: Summarize timeout
• fix(memos-local-openclaw): detect Anthropic provider in OpenClaw fallback config #1263: Anthropic provider detection (closed)