[wip] test Oauth provider revamp

AGENTS.md

@@ -138,9 +138,9 @@ Dexto infers its execution environment to enable context-aware defaults and path
 Dexto’s supported models live in core and are primarily sourced from `models.dev`.
 
 - **Registry source of truth:** `packages/core/src/llm/registry/index.ts` (consumes the generated snapshot + any manual overlays).
-- **Generated snapshot:** `packages/core/src/llm/registry/models.generated.ts` (generated from `models.dev` via `scripts/sync-llm-registry.ts`).
-    - Update: `pnpm run sync-llm-registry`
-    - Verify clean repo (CI-style): `pnpm run sync-llm-registry:check`
+- **Generated snapshot:** `packages/core/src/llm/registry/models.generated.ts` (generated from `models.dev` via `scripts/sync-llm-models.ts`).
+    - Update: `pnpm run sync-llm-models`
+    - Verify clean repo (CI-style): `pnpm run sync-llm-models:check`
 - **Gateway transform validation:** `packages/core/src/llm/registry/index.test.ts` includes a full-sweep check that our native→OpenRouter ID transform still matches the committed OpenRouter catalog snapshot (runs in `pnpm test`; catches naming drift like Anthropic dotted versions or Gemini `-001`).
 - **Manual overlays / missing models:** `packages/core/src/llm/registry/models.manual.ts` (e.g. models missing capability metadata upstream).
 - **Curation for UI/onboarding:** `packages/core/src/llm/curation-config.ts` (explicit curated model IDs; used by `/llm/catalog?scope=curated` and default pickers).

feature-plans/oauth-and-more-providers/NPM_TRANSPORT_COVERAGE.md

@@ -0,0 +1,299 @@
+# NPM / Transport Coverage Snapshot (models.dev → runtime mapping)
+
+Date: **2026-02-18**
+
+Source: `https://models.dev/api.json`
+
+Purpose: models.dev encodes the intended Vercel AI SDK provider module via `provider.npm`. Grouping provider IDs by `npm` gives us a *small, stable set of “transport kinds”* to implement, while still enabling a path to supporting “all providers”.
+
+This file is a concrete snapshot of:
+- models.dev provider IDs grouped by `npm` (23 unique values as of this date)
+- what Dexto can support *immediately* (based on existing core dependencies + transports)
+- what requires additional work (new deps and/or bespoke drivers)
+
+Related:
+- Provider IDs snapshot: [`PROVIDER_COVERAGE.md`](./PROVIDER_COVERAGE.md)
+- Direction rationale: [`UPDATED_DIRECTION.md`](./UPDATED_DIRECTION.md)
+
+---
+
+## 1) Summary
+
+- Total models.dev providers: **94**
+- Unique `provider.npm` values: **23**
+- Providers covered by Dexto’s **existing** transport surface (no new deps): **81 / 94**
+  - This is because **64 / 94** providers are `@ai-sdk/openai-compatible` (OpenAI Chat + baseURL).
+
+Unique npm counts (from models.dev):
+- 64 `@ai-sdk/openai-compatible`
+- 7 `@ai-sdk/anthropic`
+- 2 `@ai-sdk/openai`
+- 2 `@ai-sdk/azure`
+- 1 each for the remaining 18 values
+
+---
+
+## 2) Dexto “transport kinds” (stable) vs provider IDs (dynamic)
+
+The intended architecture (post Phase 1.5/2.3):
+- **Transport kinds** in core remain a small, stable set (OpenAI, OpenAI-compatible, Anthropic, Vertex, Bedrock, etc.).
+- **Provider IDs** can expand dynamically (models.dev’s 94 providers) by mapping:
+  - `provider.npm` → transport kind
+  - `provider.api` (when present) → baseURL
+  - `provider.env` / `provider.doc` → connect UX hints
+
+This keeps a clear path to “all providers” while keeping runtime code small; it’s fine if `LLM_PROVIDERS` becomes a 94-item union **as long as it’s generated** (not hand-maintained).
+
+---
+
+## 3) Coverage status (as of this repo today)
+
+### 3.1 Covered by existing core dependencies (81 providers)
+
+These `npm` values already correspond to SDKs we have (or can route via `createOpenAI`), so Phase 2.3 can expose these providers without adding new packages:
+
+- `@ai-sdk/openai-compatible` (64) → `createOpenAI({ baseURL }).chat(model)`
+- `@ai-sdk/openai` (2) → `createOpenAI({ baseURL? }).responses(model)` (or `.chat` where needed)
+- `@ai-sdk/anthropic` (7) → `createAnthropic({ baseURL? })(model)`
+- `@ai-sdk/google` (1) → `createGoogleGenerativeAI({ apiKey })(model)`
+- `@ai-sdk/google-vertex` (1) → `createVertex({ project, location })(model)` (ADC)
+- `@ai-sdk/google-vertex/anthropic` (1) → `createVertexAnthropic({ project, location })(model)` (ADC)
+- `@ai-sdk/groq` (1) → `createGroq({ apiKey })(model)`
+- `@ai-sdk/xai` (1) → `createXai({ apiKey })(model)`
+- `@ai-sdk/cohere` (1) → `createCohere({ apiKey })(model)`
+- `@ai-sdk/amazon-bedrock` (1) → `createAmazonBedrock({ region })(model)` (AWS creds)
+- `@openrouter/ai-sdk-provider` (1) → Dexto can route via `createOpenAI({ baseURL: openrouter }).responses(model)` (no need to import the OpenRouter SDK initially)
+
+### 3.2 Not covered yet (13 providers; 12 npm values)
+
+These providers require either:
+- adding the provider SDK package + wiring, or
+- a deliberate compatibility strategy (eg treat as OpenAI-compatible by adding a curated baseURL if the upstream supports it), or
+- deferring support.
+
+Not covered `npm` values:
+- `@ai-sdk/azure` (2 providers)
+- `@ai-sdk/cerebras` (1)
+- `@ai-sdk/deepinfra` (1)
+- `@ai-sdk/mistral` (1)
+- `@ai-sdk/perplexity` (1)
+- `@ai-sdk/togetherai` (1)
+- `@ai-sdk/gateway` (1) — Vercel AI Gateway provider module in models.dev
+- `@ai-sdk/vercel` (1) — v0 provider module
+- `ai-gateway-provider` (1) — Cloudflare AI Gateway provider module
+- `venice-ai-sdk-provider` (1)
+- `@gitlab/gitlab-ai-provider` (1)
+- `@jerome-benoit/sap-ai-provider-v2` (1)
+
+---
+
+## 4) Provider IDs grouped by `provider.npm` (snapshot)
+
+## @ai-sdk/amazon-bedrock (1)
+
+```text
+amazon-bedrock
+```
+
+## @ai-sdk/anthropic (7)
+
+```text
+anthropic
+kimi-for-coding
+minimax
+minimax-cn
+minimax-cn-coding-plan
+minimax-coding-plan
+zenmux
+```
+
+## @ai-sdk/azure (2)
+
+```text
+azure
+azure-cognitive-services
+```
+
+## @ai-sdk/cerebras (1)
+
+```text
+cerebras
+```
+
+## @ai-sdk/cohere (1)
+
+```text
+cohere
+```
+
+## @ai-sdk/deepinfra (1)
+
+```text
+deepinfra
+```
+
+## @ai-sdk/gateway (1)
+
+```text
+vercel
+```
+
+## @ai-sdk/google (1)
+
+```text
+google
+```
+
+## @ai-sdk/google-vertex (1)
+
+```text
+google-vertex
+```
+
+## @ai-sdk/google-vertex/anthropic (1)
+
+```text
+google-vertex-anthropic
+```
+
+## @ai-sdk/groq (1)
+
+```text
+groq
+```
+
+## @ai-sdk/mistral (1)
+
+```text
+mistral
+```
+
+## @ai-sdk/openai (2)
+
+```text
+openai
+vivgrid
+```
+
+## @ai-sdk/openai-compatible (64)
+
+```text
+302ai
+abacus
+aihubmix
+alibaba
+alibaba-cn
+bailing
+baseten
+berget
+chutes
+cloudferro-sherlock
+cloudflare-workers-ai
+cortecs
+deepseek
+fastrouter
+fireworks-ai
+firmware
+friendli
+github-copilot
+github-models
+helicone
+huggingface
+iflowcn
+inception
+inference
+io-net
+jiekou
+kilo
+kuae-cloud-coding-plan
+llama
+lmstudio
+lucidquery
+moark
+modelscope
+moonshotai
+moonshotai-cn
+morph
+nano-gpt
+nebius
+nova
+novita-ai
+nvidia
+ollama-cloud
+opencode
+ovhcloud
+poe
+privatemode-ai
+qihang-ai
+qiniu-ai
+requesty
+scaleway
+siliconflow
+siliconflow-cn
+stackit
+stepfun
+submodel
+synthetic
+upstage
+vultr
+wandb
+xiaomi
+zai
+zai-coding-plan
+zhipuai
+zhipuai-coding-plan
+```
+
+## @ai-sdk/perplexity (1)
+
+```text
+perplexity
+```
+
+## @ai-sdk/togetherai (1)
+
+```text
+togetherai
+```
+
+## @ai-sdk/vercel (1)
+
+```text
+v0
+```
+
+## @ai-sdk/xai (1)
+
+```text
+xai
+```
+
+## @gitlab/gitlab-ai-provider (1)
+
+```text
+gitlab
+```
+
+## @jerome-benoit/sap-ai-provider-v2 (1)
+
+```text
+sap-ai-core
+```
+
+## @openrouter/ai-sdk-provider (1)
+
+```text
+openrouter
+```
+
+## ai-gateway-provider (1)
+
+```text
+cloudflare-ai-gateway
+```
+
+## venice-ai-sdk-provider (1)
+
+```text
+venice
+```