feat: unify conversational engine across interfaces

.github/scripts/detect_validate_surfaces.py

@@ -37,7 +37,14 @@
     "emitters/rust.py",
     "emitters/sql.py",
     "emitters/typescript.py",
+    "llm/conversation_plan.py",
+    "llm/conversation_planner.py",
+    "llm/conversation_backend.py",
+    "llm/conversation_engine.py",
     "llm/context.py",
+    "llm/provider_types.py",
+    "llm/workspace_editor.py",
+    "llm/workspace_query.py",
     "registry/__init__.py",
     "registry/resolver.py",
     "registry/signature.py",

cli/pyproject.toml

@@ -79,6 +79,7 @@ asyncio_default_test_loop_scope = "module"
 addopts = "-n auto"
 markers = [
     "docker: requires Docker daemon and internet access (opt-in via MODELABLE_DOCKER_TESTS=1)",
+    "ollama: requires a developer-controlled local Ollama server and model (opt-in via MODELABLE_OLLAMA_TESTS=1)",
 ]
 
 [tool.coverage.run]

cli/scripts/build_browser_wheel.py

@@ -49,7 +49,14 @@
     "emitters/rust.py",
     "emitters/sql.py",
     "emitters/typescript.py",
+    "llm/conversation_plan.py",
+    "llm/conversation_planner.py",
+    "llm/conversation_backend.py",
+    "llm/conversation_engine.py",
     "llm/context.py",
+    "llm/provider_types.py",
+    "llm/workspace_editor.py",
+    "llm/workspace_query.py",
     "registry/__init__.py",
     "registry/resolver.py",
     "registry/signature.py",

cli/src/modelable/browser/__init__.py

@@ -1,4 +1,5 @@
 from modelable.browser.api import BrowserCompiler
+from modelable.browser.conversation import BrowserConversationReply, BrowserConversationService
 from modelable.browser.dispatch import dispatch_browser_request
 from modelable.browser.dto import (
     BrowserArtifact,
@@ -17,6 +18,8 @@
     "BrowserCompileResult",
     "BrowserCompiler",
     "BrowserCompletionResult",
+    "BrowserConversationReply",
+    "BrowserConversationService",
     "BrowserDiagnostic",
     "BrowserFormatResult",
     "BrowserHoverResult",

cli/src/modelable/browser/api.py

@@ -2,18 +2,8 @@
 
 from types import MappingProxyType
 
-from modelable.browser.ai import (
-    build_explain_request,
-    build_generate_entity_request,
-    build_suggest_projection_request,
-    parse_explain_result,
-    parse_generate_result,
-)
 from modelable.browser.compatibility import build_browser_compatibility
 from modelable.browser.dto import (
-    BrowserAiExplainResult,
-    BrowserAiGenerateResult,
-    BrowserAiPendingResult,
     BrowserArtifact,
     BrowserCompatibilityResult,
     BrowserCompileResult,
@@ -117,6 +107,11 @@ def _load_workspace(sources: tuple[BrowserSource, ...]) -> Workspace | tuple[Bro
 class BrowserCompiler:
     def __init__(self) -> None:
         self.language = LanguageWorkspace()
+        self._sources: tuple[BrowserSource, ...] = ()
+
+    @property
+    def sources(self) -> tuple[BrowserSource, ...]:
+        return self._sources
 
     def open_workspace(
         self,
@@ -130,6 +125,7 @@ def open_workspace(
             workspace_revision,
             tuple(LanguageDocument.from_text(source.uri, source.text, source.version) for source in sources),
         )
+        self._sources = sources
         return BrowserWorkspaceResult(
             workspace_revision=synchronization.revision,
             diagnostics=tuple(_browser_diagnostic(diagnostic) for diagnostic in synchronization.diagnostics),
@@ -277,55 +273,6 @@ def governance(
             raise BrowserLanguageError("LANGUAGE_UNAVAILABLE")
         return build_browser_governance(semantic, workspace_revision)
 
-    def ai_generate(
-        self,
-        workspace_revision: int,
-        action: str,
-        parameters: dict[str, object],
-        llm_response_content: str | None,
-    ) -> BrowserAiPendingResult | BrowserAiGenerateResult:
-        if workspace_revision != self.language.revision:
-            raise BrowserLanguageError("STALE_WORKSPACE")
-
-        if llm_response_content is None:
-            if action == "generate_entity":
-                return build_generate_entity_request(
-                    self.language,
-                    description=str(parameters.get("description", "")),
-                    domain_name=parameters.get("domainName"),  # type: ignore[arg-type]
-                    model_name=parameters.get("modelName"),  # type: ignore[arg-type]
-                )
-            if action == "suggest_projection":
-                return build_suggest_projection_request(
-                    self.language,
-                    source_ref=str(parameters.get("sourceRef", "")),
-                    consumer_domain=str(parameters.get("consumerDomain", "")),
-                )
-            raise BrowserRequestValidationError(f"Unknown generate action: {action}")
-
-        return parse_generate_result(llm_response_content)
-
-    def ai_explain(
-        self,
-        workspace_revision: int,
-        parameters: dict[str, object],
-        llm_response_content: str | None,
-    ) -> BrowserAiPendingResult | BrowserAiExplainResult:
-        if workspace_revision != self.language.revision:
-            raise BrowserLanguageError("STALE_WORKSPACE")
-
-        if llm_response_content is None:
-            ref = parameters.get("ref")
-            raw_index = parameters.get("diagnosticIndex")
-            diag_index: int | None = int(raw_index) if isinstance(raw_index, (int, float, str)) else None
-            return build_explain_request(
-                self.language,
-                ref=str(ref) if ref is not None else None,
-                diagnostic_index=diag_index,
-            )
-
-        return parse_explain_result(llm_response_content)
-
     def _validate_language_request(
         self,
         request: BrowserLanguagePosition,