fix: add community config endpoint and fix provider routing

[neuromechanist]

Summary

Fixes two critical issues preventing the Qwen model from working:

1. Missing config endpoint: Widget was calling GET /communities/{communityId} but that endpoint didn't exist
2. Wrong provider format: Backend used {"only": [provider]} but OpenRouter requires {"order": [provider]}

Changes

Backend

• Add GET /{community_id}/ endpoint to return community configuration (id, name, description, default_model, default_model_provider)
• Fix OpenRouter provider routing in litellm_llm.py: change "only" to "order"

Frontend

• Update widget to call /{communityId}/ instead of /communities/{communityId}
• Remove hardcoded fallback model (openai/gpt-oss-120b)
• Show proper error messages when config fetch fails

Documentation

• Add .context/api-structure.md - comprehensive API implementation guide
  • Community-based routing pattern
  • Model selection logic
  • Common mistakes and how to avoid them
  • Testing recipes
• Reorganize CLAUDE.md Key Documentation section by purpose

Testing

Tested locally:

• GET /hed/ returns correct config with qwen/qwen3-235b-a22b-2507
• OpenRouter API accepts {"order": ["DeepInfra/FP8"]} (tested with curl)

Impact

• Widget will now show correct default model in settings
• Backend will successfully route to DeepInfra/FP8 provider for Qwen model
• Prevents similar routing/endpoint mistakes in the future

Closes #93

[github-actions]

🚀 Preview Deployment

Name	Link
Preview URL	https://feature-fix-qwen-provider-is.osa-demo.pages.dev
Branch	feature/fix-qwen-provider-issue
Commit	ba040d1

This preview will be updated automatically when you push new commits.

[neuromechanist]

PR Review Summary

Overall Assessment: Conditional Approval - Core fixes are correct, but widget needs better error handling before merge.

Issues Found

Priority	Issue	Location	Description
🔴 CRITICAL	Widget silent failure	frontend/osa-chat-widget.js:1353-1390	Widget continues operating when config fetch fails, leaving communityDefaultModel = null. Should disable widget completely.
🔴 CRITICAL	Settings masks error	frontend/osa-chat-widget.js:1408	Shows "Default (Community Setting)" when model is null, hiding the failure. Should show "ERROR: Not configured" and disable option.
🔴 CRITICAL	Hint hides error	frontend/osa-chat-widget.js:1432-1434	Model hint silently omitted when config fails. Should show red error text.
🟡 HIGH	Broad exception handling	src/api/routers/community.py:803-813	Catches all exceptions with generic "Internal server error". Should use specific handlers for auth errors, rate limits, timeouts.
🔵 MINOR	Port number clarity	.context/api-structure.md:314	Testing examples use dev port (38529) but default is prod (38528). Add clarifying comment.
🔵 MINOR	Missing validation	src/api/routers/community.py:949-974	Config endpoint doesn't validate default_model is non-null. Add validation before return.

What's Good ✅

• ✅ Core fixes are correct (provider routing, config endpoint)
• ✅ Excellent documentation with accurate examples
• ✅ Follows CLAUDE.md guidelines (no hardcoded fallbacks)
• ✅ All code examples tested and working

Recommendation

Fix the 3 critical widget issues before merging:

1. Disable widget when config fetch fails (add disableWidget() function)
2. Show error state in settings dropdown when model is null
3. Show error state in model hint when model is null

The high-priority backend exception handling can be addressed in a follow-up PR as it's a pre-existing pattern across endpoints.

---

Review completed by: code-reviewer, comment-analyzer, silent-failure-hunter agents

[neuromechanist]

✅ All Critical Issues Addressed

Fixes Applied

Issue	Status	Commit
🔴 Widget silent failure	✅ FIXED	b4ba858 - Added disableWidget() function that disables input/send button on config errors
🔴 Settings masks error	✅ FIXED	b4ba858 - Shows "ERROR: Not configured" and disables dropdown option when model is null
🔴 Hint hides error	✅ FIXED	b4ba858 - Shows red error text when model is null
🔵 Port clarity	✅ FIXED	335a269 - Added comment clarifying dev vs prod port
🔵 Missing validation	✅ FIXED	335a269 - Config endpoint validates default_model before return

High-Priority Issue

🟡 Broad exception handling - Deferred to follow-up PR as it's a pre-existing pattern across all endpoints

---

Status: Ready to merge ✅

All critical widget error handling issues have been resolved. The widget now:

• Disables itself completely when config fetch fails
• Shows clear error states in all UI elements (dropdown, hint text)
• Provides actionable error messages to users

The provider routing fix ("only" → "order") and new config endpoint will work once deployed to backend.