feat(ai): phase B function calling in-app (agent loop + endpoint /agent)

- backend/ai_chat.py: chat_completion provider-agnostique (OpenAI-compat tools/tool_calls + Gemini functionDeclarations/functionCall), retry sans tools si rejete
- backend/agent/loop.py: run_agent multi-etapes (limite 10, truncation, confirmation two-step), LLM injectable
- endpoint opt-in POST /api/ai/bookslm/agent (events SSE tool/message/confirmation), extraction _resolve_system_prompt
- tests: test_ai_chat.py, test_agent_loop.py + 3 tests endpoint (728 passed au total)
- ROADMAP B1/B2/B3/B7 livres ; B4/B5/B6 restants

tests/test_bookslm.py

@@ -746,3 +746,80 @@ class TestBooksLMChatModes:
         )
         assert resp.status_code in (200, 500), f"unexpected {resp.status_code}: {resp.text[:200]}"
         assert resp.status_code != 422
+
+
+# ── Tool-calling agent endpoint ───────────────────────────────────────
+
+
+class TestBooksLMAgentEndpoint:
+    """Tests for POST /api/ai/bookslm/agent (Phase B)."""
+
+    def test_agent_requires_auth(self, bookslm_client):
+        resp = bookslm_client.post(
+            "/api/ai/bookslm/agent",
+            json={"vault": "TestVault", "directory": "", "message": "Hello"},
+        )
+        assert resp.status_code == 401
+
+    def test_agent_tool_call_flow(self, bookslm_client, monkeypatch):
+        import backend.bookslm_routes as routes
+        from backend.ai_chat import LLMResponse, ToolCall
+
+        responses = [
+            LLMResponse(tool_calls=[ToolCall(id="1", name="list_directory", arguments={"vault": "TestVault"})]),
+            LLMResponse(content="Il y a des fichiers."),
+        ]
+
+        async def fake_chat_completion(messages, **kwargs):
+            return responses.pop(0)
+
+        monkeypatch.setattr(routes, "chat_completion", fake_chat_completion)
+        monkeypatch.setattr(routes, "_resolve_provider_name", lambda requested: "deepseek")
+
+        token, _ = _login_bookslm(bookslm_client)
+        resp = bookslm_client.post(
+            "/api/ai/bookslm/agent",
+            json={"vault": "TestVault", "directory": "", "message": "liste les fichiers", "mode": "directory"},
+            headers={"Authorization": f"Bearer {token}"},
+        )
+        assert resp.status_code == 200
+        body = resp.text
+        assert "event: tool" in body
+        assert "list_directory" in body
+        assert "event: message" in body
+        assert "Il y a des fichiers." in body
+
+    def test_agent_pauses_for_confirmation(self, bookslm_client, monkeypatch):
+        import backend.bookslm_routes as routes
+        from backend.ai_chat import LLMResponse, ToolCall
+        from backend.tools import registry
+        from backend.tools.api import ToolRisk
+        from backend.tools.registry import ToolSpec
+        from backend.tools.schemas import ListVaultsInput
+
+        spec = ToolSpec(
+            name="_agent_write",
+            description="write for tests",
+            input_model=ListVaultsInput,
+            handler=lambda ctx, params: {"ok": True},
+            risk=ToolRisk.WRITE,
+        )
+        monkeypatch.setitem(registry._REGISTRY, "_agent_write", spec)
+
+        responses = [LLMResponse(tool_calls=[ToolCall(id="1", name="_agent_write", arguments={})])]
+
+        async def fake_chat_completion(messages, **kwargs):
+            return responses.pop(0)
+
+        monkeypatch.setattr(routes, "chat_completion", fake_chat_completion)
+        monkeypatch.setattr(routes, "_resolve_provider_name", lambda requested: "deepseek")
+
+        token, _ = _login_bookslm(bookslm_client)
+        resp = bookslm_client.post(
+            "/api/ai/bookslm/agent",
+            json={"vault": "TestVault", "directory": "", "message": "crée un fichier", "mode": "directory"},
+            headers={"Authorization": f"Bearer {token}"},
+        )
+        assert resp.status_code == 200
+        assert "event: confirmation" in resp.text
+        assert "_agent_write" in resp.text