feat(llm): migrate providers to ai sdk (#1449)

* feat(llm): migrate runtime to ai sdk

* fix(llm): harden runtime edge cases

* fix(llm): harden tool map keys

* fix(llm): continue legacy tool loops

* fix(llm): merge allof required keys

* feat(llm): migrate azure runtime

* fix(llm): harden ai sdk adapters

* refactor(provider): retire legacy runtime

* refactor(provider): simplify provider layer

Author: zerob13

docs/README.md

@@ -1,6 +1,6 @@
 # DeepChat 文档索引
 
-本文档反映 `2026-03-23` 完成的 legacy `AgentPresenter` retirement 之后的代码结构。
+本文档反映 `2026-04-11` 完成 legacy `AgentPresenter` retirement 与 legacy provider runtime retirement 之后的代码结构。
 
 当前聊天主链路已经收敛为：
 
@@ -36,6 +36,13 @@ Renderer
 | [docs/specs/legacy-agentpresenter-retirement/spec.md](./specs/legacy-agentpresenter-retirement/spec.md) | 本次 retirement 的目标、范围、兼容边界 |
 | [docs/specs/legacy-agentpresenter-retirement/plan.md](./specs/legacy-agentpresenter-retirement/plan.md) | 迁移/归档/验证计划 |
 | [docs/specs/legacy-agentpresenter-retirement/tasks.md](./specs/legacy-agentpresenter-retirement/tasks.md) | 已执行清单 |
+| [docs/specs/legacy-llm-provider-runtime-retirement/spec.md](./specs/legacy-llm-provider-runtime-retirement/spec.md) | legacy provider runtime retirement 规格 |
+| [docs/specs/legacy-llm-provider-runtime-retirement/plan.md](./specs/legacy-llm-provider-runtime-retirement/plan.md) | provider runtime 收口与依赖清理计划 |
+| [docs/specs/legacy-llm-provider-runtime-retirement/tasks.md](./specs/legacy-llm-provider-runtime-retirement/tasks.md) | provider runtime 退役执行清单 |
+| [docs/specs/provider-layer-simplification/spec.md](./specs/provider-layer-simplification/spec.md) | provider layer 第二轮内部收口规格 |
+| [docs/specs/provider-layer-simplification/plan.md](./specs/provider-layer-simplification/plan.md) | registry + generic provider 合并计划 |
+| [docs/specs/provider-layer-simplification/tasks.md](./specs/provider-layer-simplification/tasks.md) | provider layer 第二轮执行清单 |
+| [docs/specs/ai-sdk-runtime/spec.md](./specs/ai-sdk-runtime/spec.md) | AI SDK runtime 规格，现已更新为 retired 状态 |
 | [docs/specs/architecture-simplification/spec.md](./specs/architecture-simplification/spec.md) | 整体减负治理规格 |
 | [docs/specs/architecture-simplification/plan.md](./specs/architecture-simplification/plan.md) | 分层/基线/guard 计划 |
 | [docs/specs/architecture-simplification/tasks.md](./specs/architecture-simplification/tasks.md) | 首期实施清单 |
@@ -62,10 +69,14 @@ docs/
 ├── specs/
 │   ├── agent-cleanup/
 │   ├── architecture-simplification/
+│   ├── ai-sdk-runtime/
+│   ├── provider-layer-simplification/
+│   ├── legacy-llm-provider-runtime-retirement/
 │   └── legacy-agentpresenter-retirement/
 └── archives/
     ├── legacy-agentpresenter-architecture.md
     ├── legacy-agentpresenter-flows.md
+    ├── legacy-llm-provider-runtime.md
     ├── thread-presenter-migration-plan.md
     └── workspace-agent-refactoring-summary.md
 ```
@@ -78,6 +89,7 @@ docs/
 | --- | --- |
 | [archives/legacy-agentpresenter-architecture.md](./archives/legacy-agentpresenter-architecture.md) | 旧 `AgentPresenter` 架构快照 |
 | [archives/legacy-agentpresenter-flows.md](./archives/legacy-agentpresenter-flows.md) | 旧 `startStreamCompletion` / permission / loop 流程 |
+| [archives/legacy-llm-provider-runtime.md](./archives/legacy-llm-provider-runtime.md) | 旧 provider runtime 的历史归档与提交锚点 |
 | [archives/thread-presenter-migration-plan.md](./archives/thread-presenter-migration-plan.md) | 历史迁移设计 |
 | [archives/workspace-agent-refactoring-summary.md](./archives/workspace-agent-refactoring-summary.md) | 历史工作区改造总结 |
 

docs/architecture/baselines/dependency-report.md

@@ -32,7 +32,6 @@ Generated on 2026-04-03.
 - `eventbus.ts`: 57
 - `presenter/index.ts`: 44
 - `presenter/llmProviderPresenter/runtimePorts.ts`: 34
-- `presenter/llmProviderPresenter/providers/openAICompatibleProvider.ts`: 24
 - `presenter/llmProviderPresenter/baseProvider.ts`: 19
 - `presenter/remoteControlPresenter/types.ts`: 18
 - `presenter/sqlitePresenter/tables/baseTable.ts`: 16
@@ -51,9 +50,7 @@ Generated on 2026-04-03.
 - `presenter/index.ts -> presenter/windowPresenter/index.ts -> presenter/windowPresenter/FloatingChatWindow.ts -> presenter/index.ts`
 - `presenter/index.ts -> presenter/shortcutPresenter.ts -> presenter/index.ts`
 - `presenter/index.ts -> presenter/llmProviderPresenter/index.ts -> presenter/llmProviderPresenter/baseProvider.ts -> presenter/devicePresenter/index.ts -> presenter/index.ts`
-- `presenter/index.ts -> presenter/llmProviderPresenter/index.ts -> presenter/llmProviderPresenter/managers/providerInstanceManager.ts -> presenter/llmProviderPresenter/providers/deepseekProvider.ts -> presenter/llmProviderPresenter/providers/openAICompatibleProvider.ts -> presenter/index.ts`
 - `presenter/index.ts -> presenter/llmProviderPresenter/index.ts -> presenter/llmProviderPresenter/managers/providerInstanceManager.ts -> presenter/llmProviderPresenter/providers/githubCopilotProvider.ts -> presenter/githubCopilotDeviceFlow.ts -> presenter/index.ts`
-- `presenter/index.ts -> presenter/llmProviderPresenter/index.ts -> presenter/llmProviderPresenter/managers/providerInstanceManager.ts -> presenter/llmProviderPresenter/providers/openAIResponsesProvider.ts -> presenter/index.ts`
 - `presenter/filePresenter/mime.ts -> presenter/filePresenter/CsvFileAdapter.ts -> presenter/filePresenter/BaseFileAdapter.ts -> presenter/filePresenter/mime.ts`
 - `presenter/index.ts -> presenter/sessionPresenter/index.ts -> presenter/index.ts`
 - `presenter/index.ts -> presenter/sessionPresenter/index.ts -> presenter/sessionPresenter/managers/conversationManager.ts -> presenter/index.ts`
@@ -115,4 +112,3 @@ Generated on 2026-04-03.
 - `components/json-viewer/JsonValue.ts -> components/json-viewer/JsonObject.ts -> components/json-viewer/JsonValue.ts`
 - `components/json-viewer/JsonArray.ts -> components/json-viewer/JsonValue.ts -> components/json-viewer/JsonArray.ts`
 - `composables/usePageCapture.example.ts -> composables/usePageCapture.example.ts`
-

docs/architecture/baselines/zero-inbound-candidates.md

@@ -6,15 +6,14 @@ These files have no in-repo importers inside their scope and need manual classif
 
 ## main
 
-- Candidate count: 16
+- Candidate count: 15
 
 - `env.d.ts`
 - `lib/system.ts`
 - `lib/terminalHelper.ts`
 - `presenter/browser/BrowserContextBuilder.ts`
 - `presenter/configPresenter/aes.ts`
 - `presenter/llmProviderPresenter/oauthHelper.ts`
-- `presenter/llmProviderPresenter/providers/openAIProvider.ts`
 - `presenter/mcpPresenter/agentMcpFilter.ts`
 - `presenter/searchPrompts/searchPrompts.ts`
 - `presenter/sessionPresenter/events.ts`
@@ -73,4 +72,3 @@ These files have no in-repo importers inside their scope and need manual classif
 - `stores/systemPromptStore.ts`
 - `utils/maxOutputTokens.ts`
 - `views/SettingsTabView.vue`
-

docs/archives/legacy-llm-provider-runtime.md

@@ -0,0 +1,38 @@
+# Legacy LLM Provider Runtime Archive
+
+## Summary
+
+DeepChat previously maintained two low-level provider runtimes under `llmProviderPresenter`:
+
+- the original provider-specific SDK implementations
+- the newer shared AI SDK runtime
+
+That rollback window is now closed. The active codebase only keeps the AI SDK runtime.
+
+## Timeline
+
+- AI SDK migration landed in commit `4c8345a7`
+- Legacy runtime retirement and dependency cleanup landed after the migration stabilized
+
+## Where To Find The Old Provider Implementation
+
+Use commit `3add4093b46f15072d5ec3a65c8097e23b4907c4` to inspect the historical provider implementation and legacy runtime code.
+
+That commit is the canonical source for:
+
+- legacy provider request code
+- legacy stream parsing branches
+- provider-specific MCP conversion APIs
+- legacy rollback-path wiring
+
+## Current State
+
+- no `DEEPCHAT_LLM_RUNTIME`
+- no `llmRuntimeMode`
+- no legacy provider SDK fallback branches in active providers
+- no provider-specific MCP conversion APIs exposed from presenters
+
+For current implementation details, read:
+
+- [docs/specs/ai-sdk-runtime/spec.md](../specs/ai-sdk-runtime/spec.md)
+- [docs/specs/legacy-llm-provider-runtime-retirement/spec.md](../specs/legacy-llm-provider-runtime-retirement/spec.md)

docs/specs/ai-sdk-runtime/plan.md

@@ -0,0 +1,10 @@
+# AI SDK Runtime Plan
+
+Status: completed, rollback retired. See [../legacy-llm-provider-runtime-retirement/plan.md](../legacy-llm-provider-runtime-retirement/plan.md).
+
+1. Introduce shared AI SDK runtime modules without changing upper-layer interfaces.
+2. Migrate OpenAI-compatible and OpenAI responses providers first.
+3. Migrate Anthropic / Gemini / Vertex / Bedrock / Ollama to the shared runtime.
+4. Keep routing providers (`new-api`, `zenmux`) as thin delegates over migrated providers.
+5. Freeze `LLMCoreStreamEvent` behavior with adapter-focused tests.
+6. Retire the rollback path and delete legacy state machines.

docs/specs/ai-sdk-runtime/spec.md

@@ -0,0 +1,93 @@
+# AI SDK Runtime Spec
+
+## Status
+
+Completed in commit `4c8345a7`.
+
+As of `2026-04-11`, the rollback path is retired. See [../legacy-llm-provider-runtime-retirement/spec.md](../legacy-llm-provider-runtime-retirement/spec.md).
+
+## Goal
+
+Unify DeepChat's low-level LLM request pipeline on Vercel AI SDK while keeping the upper-layer contracts unchanged:
+
+- `BaseLLMProvider`
+- `LLMProviderPresenter`
+- `LLMCoreStreamEvent`
+- existing provider IDs, model configs, and conversation history
+
+The AI SDK runtime is the only remaining implementation.
+
+## Non-Negotiable Compatibility
+
+- No functional regression in text streaming, reasoning streaming, tool call streaming, image output, prompt cache, proxy handling, request tracing, routing, and embeddings.
+- `LLMCoreStreamEvent` event names, field names, and stop reasons remain unchanged.
+- Existing `function_call_record` history must stay reusable across providers.
+- Existing provider list / model list / provider check / key status responsibilities remain in provider classes.
+
+## Runtime Mode
+
+- Single runtime: `ai-sdk`
+- `DEEPCHAT_LLM_RUNTIME` has been removed
+- config setting `llmRuntimeMode` has been removed
+
+## Scope
+
+Shared runtime under `src/main/presenter/llmProviderPresenter/aiSdk/` provides:
+
+- provider factory
+- model / message mapper
+- MCP tool mapper
+- streaming adapter
+- image runtime
+- embedding runtime
+- provider-options mapper
+- reasoning middleware
+- legacy function-call compatibility middleware
+
+## Provider Rollout
+
+Phase 1:
+
+- `OpenAICompatibleProvider`
+- `OpenAIResponsesProvider`
+- all `extends OpenAICompatibleProvider` providers
+
+Phase 2:
+
+- `AnthropicProvider`
+- `GeminiProvider`
+- `VertexProvider`
+- `AwsBedrockProvider`
+- `OllamaProvider`
+
+Phase 3:
+
+- `NewApiProvider`
+- `ZenmuxProvider`
+
+Out of scope for first unification pass:
+
+- `AcpProvider`
+- `VoiceAIProvider`
+
+## Validation Matrix
+
+- pure text
+- reasoning native
+- reasoning via `<think>`
+- native tool streaming
+- legacy `<function_call>` fallback
+- multi-tool history replay
+- image input
+- image output
+- usage mapping
+- prompt cache mapping
+- proxy / trace / abort
+- embeddings
+- retired rollback path verification
+
+## Legacy Removal Exit Criteria
+
+- AI SDK runtime passes the provider regression matrix
+- duplicated legacy stream parsers / tool parsers have no remaining callers
+- retirement is documented in [../legacy-llm-provider-runtime-retirement/spec.md](../legacy-llm-provider-runtime-retirement/spec.md)

docs/specs/ai-sdk-runtime/tasks.md

@@ -0,0 +1,17 @@
+# AI SDK Runtime Tasks
+
+- [x] Freeze migration scope in SDD docs.
+- [x] Add shared AI SDK runtime modules. (implemented in 4c8345a7)
+- [x] Integrate OpenAI-compatible runtime path. (implemented in 4c8345a7)
+- [x] Integrate OpenAI responses runtime path. (implemented in 4c8345a7)
+- [x] Integrate Anthropic runtime path. (implemented in 4c8345a7)
+- [x] Integrate Gemini runtime path. (implemented in 4c8345a7)
+- [x] Integrate Vertex runtime path. (implemented in 4c8345a7)
+- [x] Integrate Bedrock runtime path. (implemented in 4c8345a7)
+- [x] Integrate Ollama runtime path. (implemented in 4c8345a7)
+- [x] Add regression tests for runtime adapter behavior. (implemented in 4c8345a7)
+- [x] Retire `DEEPCHAT_LLM_RUNTIME` and `llmRuntimeMode`.
+- [x] Remove legacy provider SDK fallback branches.
+- [x] Remove provider-specific MCP conversion interfaces from presenter ports.
+- [x] Record retirement history in [../legacy-llm-provider-runtime-retirement/tasks.md](../legacy-llm-provider-runtime-retirement/tasks.md).
+- [x] Run format, i18n, lint, and targeted tests.

docs/specs/legacy-llm-provider-runtime-retirement/plan.md

@@ -0,0 +1,20 @@
+# Legacy Provider Runtime Retirement Plan
+
+## Outcome
+
+Legacy provider runtime retirement is complete and no rollback path remains in the codebase.
+
+## Executed Plan
+
+1. Remove runtime selection and make AI SDK the only request runtime.
+2. Collapse provider implementations onto shared AI SDK helpers.
+3. Remove legacy MCP tool conversion surface from presenter interfaces.
+4. Delete obsolete provider SDK dependencies and refresh lockfiles.
+5. Archive the migration history and point readers to the final legacy-code commit.
+
+## Exit Conditions
+
+- No remaining source imports of `openai`, `@anthropic-ai/sdk`, `@google/genai`, `together-ai`, or `@aws-sdk/client-bedrock-runtime`
+- No `DEEPCHAT_LLM_RUNTIME` or `llmRuntimeMode` references remain
+- Main-process provider tests validate AI SDK-only behavior
+- Documentation explicitly marks the rollback path as retired

docs/specs/legacy-llm-provider-runtime-retirement/spec.md

@@ -0,0 +1,45 @@
+# Legacy Provider Runtime Retirement Spec
+
+## Status
+
+Completed on `2026-04-11`.
+
+The hidden rollback window is closed. `llmProviderPresenter` now runs on AI SDK only.
+
+## Goal
+
+Retire the legacy provider runtime and remove obsolete provider SDK dependencies without changing upper-layer contracts:
+
+- `BaseLLMProvider`
+- `LLMProviderPresenter`
+- `LLMCoreStreamEvent`
+- existing provider IDs, model configs, conversation history, and `function_call_record` compatibility
+
+## Scope
+
+- Remove `DEEPCHAT_LLM_RUNTIME`
+- Remove config key `llmRuntimeMode`
+- Delete `src/main/presenter/llmProviderPresenter/aiSdk/runtimeMode.ts`
+- Remove legacy-only provider branches, stream parsers, and MCP conversion ports
+- Keep provider-managed responsibilities that still matter:
+  - `ollama` local model management
+  - `@aws-sdk/client-bedrock` model discovery
+
+## Runtime State After Retirement
+
+- Single runtime: `ai-sdk`
+- No hidden fallback
+- No provider-specific MCP conversion APIs exposed from presenters
+- Vendor-specific request body customization is handled via AI SDK provider options mapping
+
+## Historical Anchors
+
+- AI SDK migration landed in commit `4c8345a7`
+- Legacy provider implementation can be inspected at commit `3add4093b46f15072d5ec3a65c8097e23b4907c4`
+
+## Compatibility Commitments
+
+- `LLMCoreStreamEvent` names and fields remain unchanged
+- Provider IDs and provider settings remain unchanged
+- Existing message history and `function_call_record` remain reusable
+- Routing providers (`new-api`, `zenmux`) stay as thin delegates over migrated providers

docs/specs/legacy-llm-provider-runtime-retirement/tasks.md

@@ -0,0 +1,17 @@
+# Legacy Provider Runtime Retirement Tasks
+
+- [x] Delete `src/main/presenter/llmProviderPresenter/aiSdk/runtimeMode.ts`
+- [x] Remove `DEEPCHAT_LLM_RUNTIME` and `llmRuntimeMode`
+- [x] Convert provider request paths to AI SDK-only implementations
+- [x] Remove provider-specific MCP tool conversion interfaces from presenter ports
+- [x] Replace legacy SDK type imports with local neutral types where still needed
+- [x] Remove obsolete provider SDK dependencies from `package.json`
+- [x] Rewrite provider tests around AI SDK runtime helpers and delegate routing
+- [x] Archive legacy runtime history and document the last legacy-code commit
+- [x] Run `pnpm install`
+- [x] Run `pnpm run format`
+- [x] Run `pnpm run i18n`
+- [x] Run `pnpm run lint`
+- [x] Run `pnpm run typecheck`
+- [x] Run targeted provider tests for the migrated AI SDK-only paths
+- [ ] Run `pnpm run test:main`