hugo/bDS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Resolve all 3 open questions in refactoring plan

Browse Source
This commit is contained in:
hugo
2026-03-01 19:11:29 +01:00
parent fe394a02b8
commit 0a79c40468
1 changed files with 5 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
OPENCODE_REFACTOR.md
View File

@@ -403,5 +403,8 @@ Domain logic only — no AI protocol code survives.
## Open Questions ## Open Questions
1. ~~**Zen baseURL paths**~~ — **RESOLVED**: `@ai-sdk/anthropic` appends `/messages`, `@ai-sdk/openai.chat()` appends `/chat/completions`. Verified from SDK source code and mock tests. 1. ~~**Zen baseURL paths**~~ — **RESOLVED**: `@ai-sdk/anthropic` appends `/messages`, `@ai-sdk/openai.chat()` appends `/chat/completions`. Verified from SDK source code and mock tests.
2. **Model listing** — Zen model list endpoint vs AI SDK model discovery. Likely keep thin HTTP for now. 2. ~~**Model listing**~~ — **RESOLVED**: AI SDK has no model listing/discovery API. Keep thin HTTP GET to `ZEN_MODELS_URL` (`/v1/models`) and `MISTRAL_MODELS_URL`. Move into `providers.ts` as a utility method alongside the registry. This is provider-agnostic already (OpenAI-compatible `/models` endpoint).
3. **DB message format** — Current `chatMessages` schema stores role/content/toolCallId/toolCalls. AI SDK `response.messages` may use a richer format. Evaluate whether to migrate schema or adapt at persistence layer. 3. ~~**DB message format**~~ — **RESOLVED**: Formats differ significantly. Adapt at persistence layer — no schema migration.
- **Current DB**: flat rows with `role` (string), `content` (string), `toolCallId` (string), `toolCalls` (JSON string of `[{name, args}]`)
- **AI SDK `ResponseMessage`**: `AssistantModelMessage | ToolModelMessage` with structured `content` arrays (`TextPart | ToolCallPart | ReasoningPart | ToolResultPart`)
- **Decision**: Serialize/deserialize in `ChatService` at the persistence boundary. On save: flatten AI SDK messages → flat DB rows. On load: reconstruct AI SDK `ModelMessage[]` from DB rows. This avoids a schema migration, keeps DB simple for UI queries, and the adapter is ~30 lines. Schema migration would buy us nothing — the UI reads these same rows directly for rendering.