diff --git a/.env.example b/.env.example
index 1cfa39aca5..2a2f4fbffd 100644
--- a/.env.example
+++ b/.env.example
@@ -619,6 +619,19 @@ REDIS_URL=redis://redis:6379                # REQUIRED for the self-host review
 #                                            #   BYOK (anthropic/openai) key; this gives self-host a local option.
 #                                            #   Unset = visual-vision falls back to BYOK-only.
 # AI_VISION_API_KEY=                         # bearer credential for AI_VISION_BASE_URL, if it requires one.
+# AI_ADVISORY_MODEL=qwen3:8b                  # local-inference model for advisory-only capabilities (#4364):
+#                                            #   slop advisory, e2e test-gen, issue planner, AI summaries -- none
+#                                            #   are gate-blocking, so none need frontier-model accuracy. Only
+#                                            #   used when AI_ADVISORY_BASE_URL is also set.
+# AI_ADVISORY_BASE_URL=                      # route the four advisory-only capabilities above to a SEPARATE
+#                                            #   openai-compatible endpoint (e.g. a local Ollama) instead of the
+#                                            #   shared frontier claude-code/codex chain. Unset = every advisory
+#                                            #   capability stays on the shared chain. Even when set, a
+#                                            #   capability only actually routes through it once the repo's
+#                                            #   .gittensory.yml settings.advisoryAiRouting also opts it in
+#                                            #   (global default + per-repo override) -- this binding alone
+#                                            #   changes nothing.
+# AI_ADVISORY_API_KEY=                       # bearer credential for AI_ADVISORY_BASE_URL, if it requires one.
 
 # --- Gittensory Orb (#1255; ALWAYS-ON fleet-calibration telemetry) ---
 # TELEMETRY NOTICE: running this self-hosted image contributes anonymized gate-calibration data to
diff --git a/.gittensory.yml.example b/.gittensory.yml.example
index 155867777e..902ac43096 100644
--- a/.gittensory.yml.example
+++ b/.gittensory.yml.example
@@ -869,6 +869,18 @@ settings:
   #   action: close                          # close is the only supported value. Default: close.
   #   message: "Custom close reason..."      # Default: null (built-in message).
 
+  # Advisory-AI routing (#4364): per-capability opt-in to the local-inference AI_ADVISORY binding for these
+  # four capabilities, none of which are ever gate-blocking, so none need frontier-model accuracy. Set a
+  # global default here (or in the shared/root config) and override per-repo; a capability only actually
+  # routes through AI_ADVISORY when BOTH this is true AND the AI_ADVISORY_BASE_URL binding is configured --
+  # absent either one, it stays on the shared frontier env.AI chain (byte-identical to before). All off by
+  # default.
+  # advisoryAiRouting:
+  #   slop: false          # gate.slop.aiAdvisory's second opinion. Default: false.
+  #   e2eTestGen: false    # Auto-generated E2E test scaffolding. Default: false.
+  #   planner: false       # The @gittensory plan / issue-planner completion. Default: false.
+  #   summaries: false     # AI summaries/rewrite text. Default: false.
+
 # Maintainer AI review tuning (`.gittensory.yml` top-level `review:` block). These knobs shape the advisory AI
 # review prompt and file selection only — gate/slop/secret-scan are unaffected.
 # review:
diff --git a/apps/gittensory-ui/public/openapi.json b/apps/gittensory-ui/public/openapi.json
index efb1b52ef2..1f5d994cb2 100644
--- a/apps/gittensory-ui/public/openapi.json
+++ b/apps/gittensory-ui/public/openapi.json
@@ -474,6 +474,164 @@
           "byProject"
         ]
       },
+      "PublicQualityMetrics": {
+        "type": "object",
+        "properties": {
+          "repoFullName": {
+            "type": "string"
+          },
+          "generatedAt": {
+            "type": "string"
+          },
+          "gate": {
+            "type": "object",
+            "properties": {
+              "blocked": {
+                "type": "number"
+              },
+              "blockedThenMerged": {
+                "type": "number"
+              },
+              "falsePositiveRate": {
+                "type": "number",
+                "nullable": true
+              },
+              "precisionPct": {
+                "type": "number",
+                "nullable": true
+              },
+              "topGateTypes": {
+                "type": "array",
+                "items": {
+                  "type": "object",
+                  "properties": {
+                    "gateType": {
+                      "type": "string"
+                    },
+                    "blocked": {
+                      "type": "number"
+                    },
+                    "blockedThenMerged": {
+                      "type": "number"
+                    },
+                    "falsePositiveRate": {
+                      "type": "number",
+                      "nullable": true
+                    },
+                    "precisionPct": {
+                      "type": "number",
+                      "nullable": true
+                    }
+                  },
+                  "required": [
+                    "gateType",
+                    "blocked",
+                    "blockedThenMerged",
+                    "falsePositiveRate",
+                    "precisionPct"
+                  ]
+                }
+              }
+            },
+            "required": [
+              "blocked",
+              "blockedThenMerged",
+              "falsePositiveRate",
+              "precisionPct",
+              "topGateTypes"
+            ]
+          },
+          "outcomes": {
+            "type": "object",
+            "properties": {
+              "merged": {
+                "type": "number"
+              },
+              "closed": {
+                "type": "number"
+              },
+              "mergeRatioPct": {
+                "type": "number",
+                "nullable": true
+              }
+            },
+            "required": [
+              "merged",
+              "closed",
+              "mergeRatioPct"
+            ]
+          },
+          "slop": {
+            "type": "object",
+            "properties": {
+              "totalResolved": {
+                "type": "number"
+              },
+              "overallMergeRate": {
+                "type": "number",
+                "nullable": true
+              },
+              "discriminates": {
+                "type": "boolean",
+                "nullable": true
+              }
+            },
+            "required": [
+              "totalResolved",
+              "overallMergeRate",
+              "discriminates"
+            ]
+          },
+          "trend": {
+            "type": "array",
+            "items": {
+              "type": "object",
+              "properties": {
+                "weekStart": {
+                  "type": "string"
+                },
+                "gateBlocked": {
+                  "type": "number"
+                },
+                "gateBlockedThenMerged": {
+                  "type": "number"
+                },
+                "gateFalsePositiveRate": {
+                  "type": "number",
+                  "nullable": true
+                },
+                "outcomesMerged": {
+                  "type": "number"
+                },
+                "outcomesClosed": {
+                  "type": "number"
+                },
+                "mergeRatioPct": {
+                  "type": "number",
+                  "nullable": true
+                }
+              },
+              "required": [
+                "weekStart",
+                "gateBlocked",
+                "gateBlockedThenMerged",
+                "gateFalsePositiveRate",
+                "outcomesMerged",
+                "outcomesClosed",
+                "mergeRatioPct"
+              ]
+            }
+          }
+        },
+        "required": [
+          "repoFullName",
+          "generatedAt",
+          "gate",
+          "outcomes",
+          "slop",
+          "trend"
+        ]
+      },
       "Advisory": {
         "type": "object",
         "properties": {
@@ -8497,6 +8655,13 @@
               "enabled"
             ]
           },
+          "regateSweepOrderMode": {
+            "type": "string",
+            "enum": [
+              "staleness",
+              "oldest-first"
+            ]
+          },
           "reviewCheckMode": {
             "type": "string",
             "enum": [
@@ -8605,6 +8770,19 @@
               "type": "string"
             }
           },
+          "copycatGateMode": {
+            "type": "string",
+            "enum": [
+              "off",
+              "warn",
+              "label",
+              "block"
+            ]
+          },
+          "copycatGateMinScore": {
+            "type": "number",
+            "nullable": true
+          },
           "gateDryRun": {
             "type": "boolean"
           },
@@ -8641,6 +8819,14 @@
               "block"
             ]
           },
+          "linkedIssueSatisfactionGateMode": {
+            "type": "string",
+            "enum": [
+              "off",
+              "advisory",
+              "block"
+            ]
+          },
           "firstTimeContributorGrace": {
             "type": "boolean"
           },
@@ -8867,6 +9053,29 @@
               "minConfidence"
             ]
           },
+          "advisoryAiRouting": {
+            "type": "object",
+            "properties": {
+              "slop": {
+                "type": "boolean"
+              },
+              "e2eTestGen": {
+                "type": "boolean"
+              },
+              "planner": {
+                "type": "boolean"
+              },
+              "summaries": {
+                "type": "boolean"
+              }
+            },
+            "required": [
+              "slop",
+              "e2eTestGen",
+              "planner",
+              "summaries"
+            ]
+          },
           "gittensorLabel": {
             "type": "string"
           },
@@ -8898,6 +9107,9 @@
           "badgeEnabled": {
             "type": "boolean"
           },
+          "publicQualityMetrics": {
+            "type": "boolean"
+          },
           "commandAuthorization": {
             "type": "object",
             "properties": {
@@ -9236,13 +9448,13 @@
           "reviewEvasionComment": {
             "type": "boolean"
           },
-          "createdAt": {
-            "type": "string",
-            "nullable": true
-          },
-          "updatedAt": {
+          "mergeTrainMode": {
             "type": "string",
-            "nullable": true
+            "enum": [
+              "off",
+              "audit",
+              "enforce"
+            ]
           },
           "screenshotTableGate": {
             "type": "object",
@@ -9279,43 +9491,12 @@
               "action"
             ]
           },
-          "regateSweepOrderMode": {
-            "type": "string",
-            "enum": [
-              "staleness",
-              "oldest-first"
-            ]
-          },
-          "publicQualityMetrics": {
-            "type": "boolean"
-          },
-          "mergeTrainMode": {
-            "type": "string",
-            "enum": [
-              "off",
-              "audit",
-              "enforce"
-            ]
-          },
-          "linkedIssueSatisfactionGateMode": {
+          "createdAt": {
             "type": "string",
-            "enum": [
-              "off",
-              "advisory",
-              "block"
-            ]
+            "nullable": true
           },
-          "copycatGateMode": {
+          "updatedAt": {
             "type": "string",
-            "enum": [
-              "off",
-              "warn",
-              "label",
-              "block"
-            ]
-          },
-          "copycatGateMinScore": {
-            "type": "number",
             "nullable": true
           }
         },
@@ -9846,11 +10027,18 @@
                   "enabled"
                 ]
               },
-              "reviewCheckMode": {
+              "regateSweepOrderMode": {
                 "type": "string",
                 "enum": [
-                  "required",
-                  "visible",
+                  "staleness",
+                  "oldest-first"
+                ]
+              },
+              "reviewCheckMode": {
+                "type": "string",
+                "enum": [
+                  "required",
+                  "visible",
                   "disabled"
                 ]
               },
@@ -9936,6 +10124,14 @@
                   "block"
                 ]
               },
+              "linkedIssueSatisfactionGateMode": {
+                "type": "string",
+                "enum": [
+                  "off",
+                  "advisory",
+                  "block"
+                ]
+              },
               "firstTimeContributorGrace": {
                 "type": "boolean"
               },
@@ -9967,6 +10163,9 @@
               "badgeEnabled": {
                 "type": "boolean"
               },
+              "publicQualityMetrics": {
+                "type": "boolean"
+              },
               "aiReviewMode": {
                 "type": "string",
                 "enum": [
@@ -10036,24 +10235,6 @@
                   "defaultAllowed",
                   "commandOverrides"
                 ]
-              },
-              "regateSweepOrderMode": {
-                "type": "string",
-                "enum": [
-                  "staleness",
-                  "oldest-first"
-                ]
-              },
-              "publicQualityMetrics": {
-                "type": "boolean"
-              },
-              "linkedIssueSatisfactionGateMode": {
-                "type": "string",
-                "enum": [
-                  "off",
-                  "advisory",
-                  "block"
-                ]
               }
             },
             "required": [
@@ -13514,164 +13695,6 @@
           "maintainerNextSteps",
           "privateSummary"
         ]
-      },
-      "PublicQualityMetrics": {
-        "type": "object",
-        "properties": {
-          "repoFullName": {
-            "type": "string"
-          },
-          "generatedAt": {
-            "type": "string"
-          },
-          "gate": {
-            "type": "object",
-            "properties": {
-              "blocked": {
-                "type": "number"
-              },
-              "blockedThenMerged": {
-                "type": "number"
-              },
-              "falsePositiveRate": {
-                "type": "number",
-                "nullable": true
-              },
-              "precisionPct": {
-                "type": "number",
-                "nullable": true
-              },
-              "topGateTypes": {
-                "type": "array",
-                "items": {
-                  "type": "object",
-                  "properties": {
-                    "gateType": {
-                      "type": "string"
-                    },
-                    "blocked": {
-                      "type": "number"
-                    },
-                    "blockedThenMerged": {
-                      "type": "number"
-                    },
-                    "falsePositiveRate": {
-                      "type": "number",
-                      "nullable": true
-                    },
-                    "precisionPct": {
-                      "type": "number",
-                      "nullable": true
-                    }
-                  },
-                  "required": [
-                    "gateType",
-                    "blocked",
-                    "blockedThenMerged",
-                    "falsePositiveRate",
-                    "precisionPct"
-                  ]
-                }
-              }
-            },
-            "required": [
-              "blocked",
-              "blockedThenMerged",
-              "falsePositiveRate",
-              "precisionPct",
-              "topGateTypes"
-            ]
-          },
-          "outcomes": {
-            "type": "object",
-            "properties": {
-              "merged": {
-                "type": "number"
-              },
-              "closed": {
-                "type": "number"
-              },
-              "mergeRatioPct": {
-                "type": "number",
-                "nullable": true
-              }
-            },
-            "required": [
-              "merged",
-              "closed",
-              "mergeRatioPct"
-            ]
-          },
-          "slop": {
-            "type": "object",
-            "properties": {
-              "totalResolved": {
-                "type": "number"
-              },
-              "overallMergeRate": {
-                "type": "number",
-                "nullable": true
-              },
-              "discriminates": {
-                "type": "boolean",
-                "nullable": true
-              }
-            },
-            "required": [
-              "totalResolved",
-              "overallMergeRate",
-              "discriminates"
-            ]
-          },
-          "trend": {
-            "type": "array",
-            "items": {
-              "type": "object",
-              "properties": {
-                "weekStart": {
-                  "type": "string"
-                },
-                "gateBlocked": {
-                  "type": "number"
-                },
-                "gateBlockedThenMerged": {
-                  "type": "number"
-                },
-                "gateFalsePositiveRate": {
-                  "type": "number",
-                  "nullable": true
-                },
-                "outcomesMerged": {
-                  "type": "number"
-                },
-                "outcomesClosed": {
-                  "type": "number"
-                },
-                "mergeRatioPct": {
-                  "type": "number",
-                  "nullable": true
-                }
-              },
-              "required": [
-                "weekStart",
-                "gateBlocked",
-                "gateBlockedThenMerged",
-                "gateFalsePositiveRate",
-                "outcomesMerged",
-                "outcomesClosed",
-                "mergeRatioPct"
-              ]
-            }
-          }
-        },
-        "required": [
-          "repoFullName",
-          "generatedAt",
-          "gate",
-          "outcomes",
-          "slop",
-          "trend"
-        ]
       }
     },
     "parameters": {},
@@ -13784,6 +13807,46 @@
         }
       }
     },
+    "/v1/public/repos/{owner}/{repo}/quality": {
+      "get": {
+        "parameters": [
+          {
+            "schema": {
+              "type": "string"
+            },
+            "required": true,
+            "name": "owner",
+            "in": "path"
+          },
+          {
+            "schema": {
+              "type": "string"
+            },
+            "required": true,
+            "name": "repo",
+            "in": "path"
+          }
+        ],
+        "responses": {
+          "200": {
+            "description": "Public per-repo review-quality metrics: gate false-positive rates, merge-vs-close ratio, and weekly trend. Aggregate counts only; opt-in via publicQualityMetrics.",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/PublicQualityMetrics"
+                }
+              }
+            }
+          },
+          "404": {
+            "description": "Repo is unknown/private/uninstalled or has not opted in"
+          },
+          "503": {
+            "description": "Public quality metrics are temporarily unavailable"
+          }
+        }
+      }
+    },
     "/v1/registry/snapshot": {
       "get": {
         "responses": {
@@ -16603,11 +16666,14 @@
         ]
       }
     },
-    "/v1/internal/jobs/refresh-scoring-model": {
+    "/v1/internal/jobs/generate-review-recap": {
       "post": {
         "responses": {
           "202": {
-            "description": "Internal job queued"
+            "description": "Maintainer review recap digest queued (#1963)"
+          },
+          "400": {
+            "description": "Missing repoFullName"
           },
           "401": {
             "description": "Invalid internal token"
@@ -16623,7 +16689,7 @@
         ]
       }
     },
-    "/v1/internal/jobs/refresh-upstream-drift": {
+    "/v1/internal/jobs/refresh-scoring-model": {
       "post": {
         "responses": {
           "202": {
@@ -16643,7 +16709,7 @@
         ]
       }
     },
-    "/v1/internal/jobs/file-upstream-drift-issues": {
+    "/v1/internal/jobs/refresh-upstream-drift": {
       "post": {
         "responses": {
           "202": {
@@ -16663,7 +16729,7 @@
         ]
       }
     },
-    "/v1/internal/jobs/build-contributor-evidence": {
+    "/v1/internal/jobs/file-upstream-drift-issues": {
       "post": {
         "responses": {
           "202": {
@@ -16683,7 +16749,7 @@
         ]
       }
     },
-    "/v1/internal/jobs/build-contributor-decision-packs": {
+    "/v1/internal/jobs/build-contributor-evidence": {
       "post": {
         "responses": {
           "202": {
@@ -16703,7 +16769,7 @@
         ]
       }
     },
-    "/v1/internal/jobs/build-burden-forecasts": {
+    "/v1/internal/jobs/build-contributor-decision-packs": {
       "post": {
         "responses": {
           "202": {
@@ -16723,7 +16789,7 @@
         ]
       }
     },
-    "/v1/internal/jobs/generate-signal-snapshots": {
+    "/v1/internal/jobs/build-burden-forecasts": {
       "post": {
         "responses": {
           "202": {
@@ -16743,7 +16809,7 @@
         ]
       }
     },
-    "/v1/internal/jobs/generate-weekly-value-report": {
+    "/v1/internal/jobs/generate-signal-snapshots": {
       "post": {
         "responses": {
           "202": {
@@ -16763,7 +16829,7 @@
         ]
       }
     },
-    "/v1/internal/jobs/repair-data-fidelity": {
+    "/v1/internal/jobs/generate-weekly-value-report": {
       "post": {
         "responses": {
           "202": {
@@ -16783,11 +16849,11 @@
         ]
       }
     },
-    "/v1/internal/bounties/import": {
+    "/v1/internal/jobs/repair-data-fidelity": {
       "post": {
         "responses": {
-          "200": {
-            "description": "Bounty snapshot imported"
+          "202": {
+            "description": "Internal job queued"
           },
           "401": {
             "description": "Invalid internal token"
@@ -16803,14 +16869,11 @@
         ]
       }
     },
-    "/v1/internal/jobs/generate-review-recap": {
+    "/v1/internal/bounties/import": {
       "post": {
         "responses": {
-          "202": {
-            "description": "Maintainer review recap digest queued (#1963)"
-          },
-          "400": {
-            "description": "Missing repoFullName"
+          "200": {
+            "description": "Bounty snapshot imported"
           },
           "401": {
             "description": "Invalid internal token"
@@ -16825,46 +16888,6 @@
           }
         ]
       }
-    },
-    "/v1/public/repos/{owner}/{repo}/quality": {
-      "get": {
-        "parameters": [
-          {
-            "schema": {
-              "type": "string"
-            },
-            "required": true,
-            "name": "owner",
-            "in": "path"
-          },
-          {
-            "schema": {
-              "type": "string"
-            },
-            "required": true,
-            "name": "repo",
-            "in": "path"
-          }
-        ],
-        "responses": {
-          "200": {
-            "description": "Public per-repo review-quality metrics: gate false-positive rates, merge-vs-close ratio, and weekly trend. Aggregate counts only; opt-in via publicQualityMetrics.",
-            "content": {
-              "application/json": {
-                "schema": {
-                  "$ref": "#/components/schemas/PublicQualityMetrics"
-                }
-              }
-            }
-          },
-          "404": {
-            "description": "Repo is unknown/private/uninstalled or has not opted in"
-          },
-          "503": {
-            "description": "Public quality metrics are temporarily unavailable"
-          }
-        }
-      }
     }
   },
   "servers": [
diff --git a/apps/gittensory-ui/src/lib/selfhost-env-reference.ts b/apps/gittensory-ui/src/lib/selfhost-env-reference.ts
index 953cc6600c..669c05a41b 100644
--- a/apps/gittensory-ui/src/lib/selfhost-env-reference.ts
+++ b/apps/gittensory-ui/src/lib/selfhost-env-reference.ts
@@ -5,6 +5,22 @@ export type SelfHostEnvReferenceRow = {
 };
 
 export const SELFHOST_ENV_REFERENCE_ROWS: SelfHostEnvReferenceRow[] = [
+  {
+    name: "AI_ADVISORY",
+    firstReference: "src/selfhost/ai.ts",
+  },
+  {
+    name: "AI_ADVISORY_API_KEY",
+    firstReference: "src/server.ts",
+  },
+  {
+    name: "AI_ADVISORY_BASE_URL",
+    firstReference: "src/server.ts",
+  },
+  {
+    name: "AI_ADVISORY_MODEL",
+    firstReference: "src/server.ts",
+  },
   {
     name: "AI_COMBINE",
     firstReference: "src/selfhost/ai.ts",
@@ -446,6 +462,10 @@ export const SELFHOST_ENV_REFERENCE_ROWS: SelfHostEnvReferenceRow[] = [
 export const SELFHOST_ENV_REFERENCE_MARKDOWN = [
   "| Name | First reference |",
   "| --- | --- |",
+  "| `AI_ADVISORY` | `src/selfhost/ai.ts` |",
+  "| `AI_ADVISORY_API_KEY` | `src/server.ts` |",
+  "| `AI_ADVISORY_BASE_URL` | `src/server.ts` |",
+  "| `AI_ADVISORY_MODEL` | `src/server.ts` |",
   "| `AI_COMBINE` | `src/selfhost/ai.ts` |",
   "| `AI_DUAL_REVIEW` | `src/selfhost/ai.ts` |",
   "| `AI_EMBED_API_KEY` | `src/server.ts` |",
diff --git a/config/examples/gittensory.full.yml b/config/examples/gittensory.full.yml
index da732e84eb..6b1fe56bc5 100644
--- a/config/examples/gittensory.full.yml
+++ b/config/examples/gittensory.full.yml
@@ -882,6 +882,18 @@ settings:
   #   action: close                          # close is the only supported value. Default: close.
   #   message: "Custom close reason..."      # Default: null (built-in message).
 
+  # Advisory-AI routing (#4364): per-capability opt-in to the local-inference AI_ADVISORY binding for these
+  # four capabilities, none of which are ever gate-blocking, so none need frontier-model accuracy. Set a
+  # global default here (or in the shared/root config) and override per-repo; a capability only actually
+  # routes through AI_ADVISORY when BOTH this is true AND the AI_ADVISORY_BASE_URL binding is configured --
+  # absent either one, it stays on the shared frontier env.AI chain (byte-identical to before). All off by
+  # default.
+  # advisoryAiRouting:
+  #   slop: false          # gate.slop.aiAdvisory's second opinion. Default: false.
+  #   e2eTestGen: false    # Auto-generated E2E test scaffolding. Default: false.
+  #   planner: false       # The @gittensory plan / issue-planner completion. Default: false.
+  #   summaries: false     # AI summaries/rewrite text. Default: false.
+
 # Maintainer AI review tuning (`.gittensory.yml` top-level `review:` block). These knobs shape the advisory AI
 # review prompt and file selection only — gate/slop/secret-scan are unaffected.
 # review:
diff --git a/packages/gittensory-engine/src/focus-manifest.ts b/packages/gittensory-engine/src/focus-manifest.ts
index 8d94069faa..19f3f9a932 100644
--- a/packages/gittensory-engine/src/focus-manifest.ts
+++ b/packages/gittensory-engine/src/focus-manifest.ts
@@ -9,6 +9,7 @@
  */
 import { parse as parseYaml } from "yaml";
 import type {
+  AdvisoryAiRoutingConfig,
   CombineStrategy,
   GatePolicyPack,
   GateRuleMode,
@@ -42,6 +43,7 @@ import {
   isUnlinkedIssueGuardrailMode,
   normalizeUnlinkedIssueGuardrailConfig,
 } from "./review/unlinked-issue-guardrail-config.js";
+import { normalizeAdvisoryAiRoutingConfig } from "./review/advisory-ai-routing-config.js";
 import {
   DEFAULT_SCREENSHOT_TABLE_GATE,
   isScreenshotTableGateAction,
@@ -385,6 +387,9 @@ export type FocusManifestSettings = Partial<
   // unlinkedIssueGuardrail above -- a manifest naming only `enabled` must not silently reset `whenLabels`/
   // `whenPaths`/`action`/`message` back to their defaults.
   screenshotTableGate?: Partial<ScreenshotTableGateConfig> | undefined;
+  // Advisory-AI routing (#4364): same sparse-partial merge reasoning -- a manifest naming only `slop` must
+  // not silently reset `e2eTestGen`/`planner`/`summaries` back to their (false) defaults.
+  advisoryAiRouting?: Partial<AdvisoryAiRoutingConfig> | undefined;
 };
 
 /** Field keys for the public review-panel rows a maintainer can show/hide via `review.fields`. */
@@ -1798,6 +1803,20 @@ function parseSettingsOverride(value: JsonValue | undefined, warnings: string[])
   } else if (r.screenshotTableGate !== undefined) {
     warnings.push(`Manifest "settings.screenshotTableGate" must be an object; ignoring it and keeping any existing policy.`);
   }
+  // Advisory-AI routing (#4364): same sparse-partial overlay contract as screenshotTableGate above -- a repo
+  // naming only `slop` must not silently reset `e2eTestGen`/`planner`/`summaries` back to their defaults.
+  if (typeof r.advisoryAiRouting === "object" && r.advisoryAiRouting !== null && !Array.isArray(r.advisoryAiRouting)) {
+    const rawRouting = r.advisoryAiRouting as Record<string, unknown>;
+    const validated = normalizeAdvisoryAiRoutingConfig(rawRouting, warnings);
+    const sparseRouting: Partial<AdvisoryAiRoutingConfig> = {};
+    if (typeof rawRouting.slop === "boolean") sparseRouting.slop = validated.slop;
+    if (typeof rawRouting.e2eTestGen === "boolean") sparseRouting.e2eTestGen = validated.e2eTestGen;
+    if (typeof rawRouting.planner === "boolean") sparseRouting.planner = validated.planner;
+    if (typeof rawRouting.summaries === "boolean") sparseRouting.summaries = validated.summaries;
+    out.advisoryAiRouting = sparseRouting;
+  } else if (r.advisoryAiRouting !== undefined) {
+    warnings.push(`Manifest "settings.advisoryAiRouting" must be an object; ignoring it and keeping any existing policy.`);
+  }
   // Contributor blacklist (#1425): `settings.contributorBlacklist` is a list of banned-login entries. Only set it
   // when at least one VALID entry survives normalization, so a malformed block never blanks the DB-configured
   // list via the resolver's `{...dbSettings, ...manifest.settings}` overlay. Normalization warnings are folded in.
diff --git a/packages/gittensory-engine/src/review/advisory-ai-routing-config.ts b/packages/gittensory-engine/src/review/advisory-ai-routing-config.ts
new file mode 100644
index 0000000000..fb4bf7aea1
--- /dev/null
+++ b/packages/gittensory-engine/src/review/advisory-ai-routing-config.ts
@@ -0,0 +1,35 @@
+import type { AdvisoryAiRoutingConfig } from "../types/manifest-deps-types.js";
+
+export const DEFAULT_ADVISORY_AI_ROUTING: AdvisoryAiRoutingConfig = {
+  slop: false,
+  e2eTestGen: false,
+  planner: false,
+  summaries: false,
+};
+
+function normalizeField(value: unknown, field: keyof AdvisoryAiRoutingConfig, warnings: string[]): boolean {
+  if (value === undefined) return DEFAULT_ADVISORY_AI_ROUTING[field];
+  if (typeof value === "boolean") return value;
+  warnings.push(`settings.advisoryAiRouting.${field} must be a boolean; using the default "${DEFAULT_ADVISORY_AI_ROUTING[field]}".`);
+  return DEFAULT_ADVISORY_AI_ROUTING[field];
+}
+
+/**
+ * Normalize a raw `.gittensory.yml settings.advisoryAiRouting` value into a typed config, fail-safe: any
+ * malformed field falls back to its own (false) default and pushes a warning rather than rejecting the
+ * whole block. Mirrors `normalizeUnlinkedIssueGuardrailConfig`'s per-field discipline.
+ */
+export function normalizeAdvisoryAiRoutingConfig(input: unknown, warnings: string[]): AdvisoryAiRoutingConfig {
+  if (input === undefined) return { ...DEFAULT_ADVISORY_AI_ROUTING };
+  if (typeof input !== "object" || input === null || Array.isArray(input)) {
+    warnings.push("settings.advisoryAiRouting must be an object; using the default (every capability off).");
+    return { ...DEFAULT_ADVISORY_AI_ROUTING };
+  }
+  const record = input as Record<string, unknown>;
+  return {
+    slop: normalizeField(record.slop, "slop", warnings),
+    e2eTestGen: normalizeField(record.e2eTestGen, "e2eTestGen", warnings),
+    planner: normalizeField(record.planner, "planner", warnings),
+    summaries: normalizeField(record.summaries, "summaries", warnings),
+  };
+}
diff --git a/packages/gittensory-engine/src/types/manifest-deps-types.ts b/packages/gittensory-engine/src/types/manifest-deps-types.ts
index 0195be3fda..f6bcaf577a 100644
--- a/packages/gittensory-engine/src/types/manifest-deps-types.ts
+++ b/packages/gittensory-engine/src/types/manifest-deps-types.ts
@@ -82,6 +82,21 @@ export type UnlinkedIssueGuardrailConfig = {
   minConfidence: number;
 };
 
+/** Per-capability opt-in to the local-inference AI_ADVISORY binding (#4364): each of these four ADVISORY-ONLY
+ *  (never gate-blocking) capabilities independently decides whether it routes through env.AI_ADVISORY (when
+ *  configured) instead of the shared frontier env.AI chain. Config-as-code only -- no DB column, resolved
+ *  purely from `.gittensory.yml` `settings.advisoryAiRouting` (global default in the shared/root manifest,
+ *  per-repo override), the same "config-as-code only" shape as unlinkedIssueGuardrail above. Every field
+ *  defaults to false: an operator must deliberately opt EACH capability in, and even then a repo only
+ *  actually routes through AI_ADVISORY when the binding itself is configured (env.AI_ADVISORY unset ⇒ every
+ *  capability stays on env.AI regardless of this config, byte-identical to before this existed). */
+export type AdvisoryAiRoutingConfig = {
+  slop: boolean;
+  e2eTestGen: boolean;
+  planner: boolean;
+  summaries: boolean;
+};
+
 export type ContributorBlacklistEntry = {
   login: string;
   /** Why the account is blocked. Free-text maintainer metadata; not published in automated close comments. */
@@ -315,6 +330,11 @@ export type RepositorySettings = {
    *  `.gittensory.yml settings.unlinkedIssueGuardrail` in private/global or per-repo config. Defaults
    *  all-off so a self-hoster opts into their own credibility-gate-farming defense. */
   unlinkedIssueGuardrail?: UnlinkedIssueGuardrailConfig | undefined;
+  /** Per-capability local-inference routing (#4364). Config-as-code only; set with `.gittensory.yml
+   *  settings.advisoryAiRouting` in shared/global or per-repo config (global default + per-repo override,
+   *  the same deep-merge precedence every other settings field uses). Defaults all-false so every advisory
+   *  capability stays on the shared frontier env.AI chain until an operator opts each one in. */
+  advisoryAiRouting?: AdvisoryAiRoutingConfig | undefined;
   publicSurface: "off" | "comment_and_label" | "comment_only" | "label_only";
   includeMaintainerAuthors: boolean;
   requireLinkedIssue: boolean;
diff --git a/src/env.d.ts b/src/env.d.ts
index beda0df325..81ca26b400 100644
--- a/src/env.d.ts
+++ b/src/env.d.ts
@@ -16,6 +16,13 @@ declare global {
      *  model). Built at boot from AI_VISION_BASE_URL/AI_VISION_MODEL. Absent ⇒ visual-vision advisory falls
      *  back to requiring a maintainer BYOK key (the only option before this binding existed). */
     AI_VISION?: Ai;
+    /** Self-host (advisory-tier, #4364): a DEDICATED local-inference provider shared by every capability that is
+     *  NEVER gate-blocking (slop advisory, e2e test-gen, issue planner, AI summaries) — separate from the review
+     *  chain, the embed provider, and the vision provider, since none of these need frontier-model accuracy.
+     *  Built at boot from AI_ADVISORY_BASE_URL/AI_ADVISORY_MODEL. Which capability actually routes through it is
+     *  config-driven (`.gittensory.yml`, global default + per-repo override), not hardcoded here. Absent ⇒ every
+     *  advisory capability falls back to `env.AI` (byte-identical to before this binding existed). */
+    AI_ADVISORY?: Ai;
     /** Self-host RAG vector adapter. Cloudflare no longer binds Vectorize for hosted reviews; the Node runtime
      *  injects Qdrant/sqlite/pg adapters here when configured. Absent ⇒ no RAG, review proceeds with no retrieved
      *  context. */
diff --git a/src/openapi/schemas.ts b/src/openapi/schemas.ts
index d9715803ca..dca25d0de2 100644
--- a/src/openapi/schemas.ts
+++ b/src/openapi/schemas.ts
@@ -714,6 +714,14 @@ export const RepositorySettingsSchema = z
         minConfidence: z.number().min(0).max(1),
       })
       .optional(),
+    advisoryAiRouting: z
+      .object({
+        slop: z.boolean(),
+        e2eTestGen: z.boolean(),
+        planner: z.boolean(),
+        summaries: z.boolean(),
+      })
+      .optional(),
     gittensorLabel: z.string(),
     blacklistLabel: z.string().nullable(),
     createMissingLabel: z.boolean(),
diff --git a/src/queue/processors.ts b/src/queue/processors.ts
index 3f91c38da8..22672b8023 100644
--- a/src/queue/processors.ts
+++ b/src/queue/processors.ts
@@ -390,6 +390,7 @@ import {
   VISUAL_VISION_SYSTEM_PROMPT,
 } from "../review/visual/visual-findings";
 import { incr } from "../selfhost/metrics";
+import { withAdvisoryAiEnv } from "../selfhost/ai";
 import {
   renderReviewingPlaceholder,
   shouldPostReviewingPlaceholder,
@@ -7738,7 +7739,7 @@ export async function runAiSlopForAdvisory(
         /* v8 ignore next -- reached only past this function's own `!args.advisory.headSha` early return, so headSha is always truthy here; the `?? null` is a type-level fallback for an unreachable branch. */
         metadata: { repoFullName: args.repoFullName, headSha: args.advisory.headSha ?? null },
       }).catch(() => undefined);
-      result = await runGittensoryAiSlopAdvisory(env, {
+      result = await runGittensoryAiSlopAdvisory(withAdvisoryAiEnv(env, args.settings.advisoryAiRouting?.slop === true), {
         repoFullName: args.repoFullName,
         prNumber: args.pr.number,
         title: args.pr.title,
@@ -11623,7 +11624,7 @@ async function runE2eTestGenerationAndDeliver(
     storedKey && (!args.settings.aiReviewProvider || args.settings.aiReviewProvider === storedKey.provider)
       ? { provider: storedKey.provider, key: storedKey.key, model: args.settings.aiReviewModel ?? storedKey.model }
       : null;
-  const result = await runGittensoryE2eTestGeneration(env, {
+  const result = await runGittensoryE2eTestGeneration(withAdvisoryAiEnv(env, args.settings.advisoryAiRouting?.e2eTestGen === true), {
     repoFullName: args.repoFullName,
     prNumber: args.pr.number,
     title: args.pr.title,
@@ -11915,7 +11916,7 @@ async function maybeProcessPlanCommand(
     return true;
   }
   const plan = await generateIssuePlan(
-    env,
+    withAdvisoryAiEnv(env, settings.advisoryAiRouting?.planner === true),
     { title: req.issue.title, body: req.issue.body },
     {
       actor: req.actor,
diff --git a/src/review/advisory-ai-routing-config.ts b/src/review/advisory-ai-routing-config.ts
new file mode 100644
index 0000000000..b63dbe356c
--- /dev/null
+++ b/src/review/advisory-ai-routing-config.ts
@@ -0,0 +1,35 @@
+import type { AdvisoryAiRoutingConfig } from "../types";
+
+export const DEFAULT_ADVISORY_AI_ROUTING: AdvisoryAiRoutingConfig = {
+  slop: false,
+  e2eTestGen: false,
+  planner: false,
+  summaries: false,
+};
+
+function normalizeField(value: unknown, field: keyof AdvisoryAiRoutingConfig, warnings: string[]): boolean {
+  if (value === undefined) return DEFAULT_ADVISORY_AI_ROUTING[field];
+  if (typeof value === "boolean") return value;
+  warnings.push(`settings.advisoryAiRouting.${field} must be a boolean; using the default "${DEFAULT_ADVISORY_AI_ROUTING[field]}".`);
+  return DEFAULT_ADVISORY_AI_ROUTING[field];
+}
+
+/**
+ * Normalize a raw `.gittensory.yml settings.advisoryAiRouting` value into a typed config, fail-safe: any
+ * malformed field falls back to its own (false) default and pushes a warning rather than rejecting the
+ * whole block. Mirrors `normalizeUnlinkedIssueGuardrailConfig`'s per-field discipline.
+ */
+export function normalizeAdvisoryAiRoutingConfig(input: unknown, warnings: string[]): AdvisoryAiRoutingConfig {
+  if (input === undefined) return { ...DEFAULT_ADVISORY_AI_ROUTING };
+  if (typeof input !== "object" || input === null || Array.isArray(input)) {
+    warnings.push("settings.advisoryAiRouting must be an object; using the default (every capability off).");
+    return { ...DEFAULT_ADVISORY_AI_ROUTING };
+  }
+  const record = input as Record<string, unknown>;
+  return {
+    slop: normalizeField(record.slop, "slop", warnings),
+    e2eTestGen: normalizeField(record.e2eTestGen, "e2eTestGen", warnings),
+    planner: normalizeField(record.planner, "planner", warnings),
+    summaries: normalizeField(record.summaries, "summaries", warnings),
+  };
+}
diff --git a/src/selfhost/ai.ts b/src/selfhost/ai.ts
index 039027176e..8e3b6aea76 100644
--- a/src/selfhost/ai.ts
+++ b/src/selfhost/ai.ts
@@ -1278,3 +1278,16 @@ export function resolveAiReviewerPlan(
   const onMerge = ON_MERGE_RULES.has(rawOnMerge) ? rawOnMerge : undefined;
   return { reviewers: names.slice(0, 2).map((model) => ({ model })), combine, onMerge };
 }
+
+/**
+ * Advisory-AI routing (#4364): return an `env` view whose `.AI` binding is `env.AI_ADVISORY` instead of the
+ * shared frontier chain, for a single ADVISORY-ONLY capability (slop, e2e test-gen, planner, summaries) that
+ * opted in via `settings.advisoryAiRouting`. A shallow spread — every other `env` field (including
+ * `AI_ADVISORY` itself, `AI_GATEWAY_ID`, the enablement flags) is untouched, so the capability's own
+ * fail-safe checks keep working unmodified. Falls back to the real `env` (unchanged `.AI`) whenever the
+ * capability didn't opt in OR the binding itself is unconfigured — byte-identical to before this existed in
+ * either case, exactly like `AI_EMBED`/`AI_VISION`'s own "absent ⇒ falls back" contract.
+ */
+export function withAdvisoryAiEnv(env: Env, useAdvisory: boolean): Env {
+  return useAdvisory && env.AI_ADVISORY ? { ...env, AI: env.AI_ADVISORY } : env;
+}
diff --git a/src/server.ts b/src/server.ts
index 0d3ff97909..934f35190c 100644
--- a/src/server.ts
+++ b/src/server.ts
@@ -508,6 +508,28 @@ async function main(): Promise<void> {
         model: process.env.AI_VISION_MODEL,
       }),
     );
+  // Dedicated advisory-tier provider (#4364): several capabilities (slop advisory, e2e test-gen, issue
+  // planner, AI summaries) are NEVER gate-blocking and share the review chain's frontier-only env.AI today
+  // purely because no cheaper alternative existed -- unlike AI_EMBED/AI_VISION, which each back a single
+  // narrow capability, this one binding is shared across all four, gated per-capability by
+  // `.gittensory.yml` (global default + per-repo override, see focus-manifest.ts) so routing stays
+  // config-driven, not hardcoded. Unset ⇒ absent ⇒ every advisory capability falls back to env.AI, byte-
+  // identical to before this binding existed.
+  const advisoryAi = process.env.AI_ADVISORY_BASE_URL
+    ? createOpenAiCompatibleAi({
+        baseUrl: process.env.AI_ADVISORY_BASE_URL,
+        apiKey: process.env.AI_ADVISORY_API_KEY ?? process.env.OPENAI_API_KEY,
+        model: process.env.AI_ADVISORY_MODEL,
+      })
+    : undefined;
+  if (advisoryAi)
+    console.log(
+      JSON.stringify({
+        event: "selfhost_advisory_provider",
+        baseUrl: process.env.AI_ADVISORY_BASE_URL,
+        model: process.env.AI_ADVISORY_MODEL,
+      }),
+    );
   // Dual-review plan (#dual-ai-combiner): resolve which provider(s) review + how to combine, attached to env
   // below so the review call site uses it. Undefined for a single provider's default review or no AI.
   const aiReviewPlan = resolveAiReviewerPlan(process.env);
@@ -615,6 +637,7 @@ async function main(): Promise<void> {
     AI: ai,
     ...(embedAi ? { AI_EMBED: embedAi as unknown as Ai } : {}),
     ...(visionAi ? { AI_VISION: visionAi as unknown as Ai } : {}),
+    ...(advisoryAi ? { AI_ADVISORY: advisoryAi as unknown as Ai } : {}),
     ...(aiReviewPlan ? { AI_REVIEW_PLAN: aiReviewPlan } : {}),
     SELFHOST_TRANSIENT_CACHE: webhookCache,
     // Qdrant takes priority; falls back to the backend's built-in vectorize (pgvector or sqlite-vec)
diff --git a/src/services/agent-orchestrator.ts b/src/services/agent-orchestrator.ts
index 0afbc0d37a..69a8cf05aa 100644
--- a/src/services/agent-orchestrator.ts
+++ b/src/services/agent-orchestrator.ts
@@ -29,6 +29,8 @@ import { buildContributorFit, buildContributorOutcomeHistory, buildContributorPr
 import { buildContributorOpenPrMonitor, type ContributorOpenPrMonitor } from "../signals/contributor-open-pr-monitor";
 import { buildLocalBranchAnalysis, findCurrentBranchPullRequest, type LocalBranchAnalysis, type LocalBranchAnalysisInput } from "../signals/local-branch";
 import { loadRepoFocusManifest } from "../signals/focus-manifest-loader";
+import { resolveRepositorySettings } from "../settings/repository-settings";
+import { withAdvisoryAiEnv } from "../selfhost/ai";
 import { withAgentActionExplanationCard } from "./agent-action-explanation-card";
 import { attachRecommendationSnapshots } from "./recommendation-snapshots";
 import type {
@@ -223,7 +225,12 @@ export async function executeAgentRun(env: Env, runId: string): Promise<AgentRun
 }
 
 async function attachPrivateAiSummary(env: Env, bundle: AgentRunBundle): Promise<AgentRunBundle> {
-  const summary = await summarizeAgentBundleWithAi(env, bundle, "private");
+  // Advisory-AI routing (#4364): this summary is never gate-blocking, so it's a routing candidate like
+  // slop/e2e-test-gen/planner. repoFullName can be absent for a cross-repo run (e.g. plan_next_work) --
+  // falls back to the plain env (byte-identical) rather than resolving settings for an empty key.
+  const repoFullName = String(bundle.run.payload.repoFullName ?? "");
+  const routeThroughAdvisory = repoFullName ? (await resolveRepositorySettings(env, repoFullName)).advisoryAiRouting?.summaries === true : false;
+  const summary = await summarizeAgentBundleWithAi(withAdvisoryAiEnv(env, routeThroughAdvisory), bundle, "private");
   if (summary.status === "disabled" || summary.status === "unavailable") return bundle;
   await updateAgentRun(env, bundle.run.id, {
     payload: {
diff --git a/src/signals/focus-manifest.ts b/src/signals/focus-manifest.ts
index 55d3147522..9e2f12a157 100644
--- a/src/signals/focus-manifest.ts
+++ b/src/signals/focus-manifest.ts
@@ -76,6 +76,7 @@ import { DEFAULT_TYPE_LABELS } from "../settings/pr-type-label";
 import { DEFAULT_LINKED_ISSUE_LABEL_PROPAGATION } from "../review/linked-issue-label-propagation";
 import { DEFAULT_LINKED_ISSUE_HARD_RULES } from "../review/linked-issue-hard-rules-config";
 import { DEFAULT_UNLINKED_ISSUE_GUARDRAIL } from "../review/unlinked-issue-guardrail-config";
+import { DEFAULT_ADVISORY_AI_ROUTING } from "../review/advisory-ai-routing-config";
 import { DEFAULT_SCREENSHOT_TABLE_GATE } from "../review/screenshot-table-gate";
 import { classifyChangedFile } from "./path-matchers";
 import {
@@ -501,6 +502,7 @@ export function resolveEffectiveSettings(
     linkedIssueHardRules: linkedIssueHardRulesOverride,
     unlinkedIssueGuardrail: unlinkedIssueGuardrailOverride,
     screenshotTableGate: screenshotTableGateOverride,
+    advisoryAiRouting: advisoryAiRoutingOverride,
     ...restManifestSettings
   } = manifest.settings;
   const effective: RepositorySettings = { ...dbSettings, ...restManifestSettings };
@@ -557,6 +559,15 @@ export function resolveEffectiveSettings(
       message: screenshotTableGateOverride.message ?? base.message,
     };
   }
+  if (advisoryAiRoutingOverride !== undefined) {
+    const base = dbSettings.advisoryAiRouting ?? DEFAULT_ADVISORY_AI_ROUTING;
+    effective.advisoryAiRouting = {
+      slop: advisoryAiRoutingOverride.slop ?? base.slop,
+      e2eTestGen: advisoryAiRoutingOverride.e2eTestGen ?? base.e2eTestGen,
+      planner: advisoryAiRoutingOverride.planner ?? base.planner,
+      summaries: advisoryAiRoutingOverride.summaries ?? base.summaries,
+    };
+  }
   applyGateConfigOverrides(effective, manifest.gate);
   // The dashboard "Require linked issue" toggle must not silently diverge from gate blocking: when the
   // boolean is on but linkedIssueGateMode is still off, treat it as a block requirement (#797).
diff --git a/src/types.ts b/src/types.ts
index e9f9c9eab7..7726bb3966 100644
--- a/src/types.ts
+++ b/src/types.ts
@@ -888,6 +888,10 @@ export type RepositorySettings = {
    *  `.gittensory.yml settings.unlinkedIssueGuardrail` in private/global or per-repo config. Defaults
    *  all-off so a self-hoster opts into their own credibility-gate-farming defense. */
   unlinkedIssueGuardrail?: UnlinkedIssueGuardrailConfig | undefined;
+  /** Per-capability local-inference routing (#4364). Config-as-code only; set with `.gittensory.yml
+   *  settings.advisoryAiRouting` in shared/global or per-repo config. Defaults all-false so every advisory
+   *  capability stays on the shared frontier env.AI chain until an operator opts each one in. */
+  advisoryAiRouting?: AdvisoryAiRoutingConfig | undefined;
   publicSurface: "off" | "comment_and_label" | "comment_only" | "label_only";
   includeMaintainerAuthors: boolean;
   requireLinkedIssue: boolean;
@@ -1172,6 +1176,18 @@ export type UnlinkedIssueGuardrailConfig = {
   minConfidence: number;
 };
 
+/** Per-capability opt-in to the local-inference AI_ADVISORY binding (#4364): each of these four ADVISORY-ONLY
+ *  (never gate-blocking) capabilities independently decides whether it routes through env.AI_ADVISORY (when
+ *  configured) instead of the shared frontier env.AI chain. Config-as-code only, `.gittensory.yml
+ *  settings.advisoryAiRouting` (global default in shared/root config, per-repo override); defaults all-false
+ *  so an operator must deliberately opt each capability in. */
+export type AdvisoryAiRoutingConfig = {
+  slop: boolean;
+  e2eTestGen: boolean;
+  planner: boolean;
+  summaries: boolean;
+};
+
 /** A blocked contributor (#1425, anti-abuse): a GitHub `login` plus optional maintainer metadata. The converged
  *  engine short-circuits a blacklisted author's PR/issue to a deterministic close ahead of any merit/CI/AI
  *  analysis. Metadata can come from private configuration and must not be echoed to public surfaces. */
diff --git a/test/unit/advisory-ai-routing-call-sites.test.ts b/test/unit/advisory-ai-routing-call-sites.test.ts
new file mode 100644
index 0000000000..815e9162cd
--- /dev/null
+++ b/test/unit/advisory-ai-routing-call-sites.test.ts
@@ -0,0 +1,90 @@
+// Regression coverage for the #4364 AI_ADVISORY per-capability routing at its real call sites (as opposed to
+// advisory-ai-routing-config.test.ts, which only covers the pure yml normalizer, and selfhost-ai.test.ts,
+// which only covers withAdvisoryAiEnv in isolation). Each test proves the FULL wire from
+// settings.advisoryAiRouting through to which env.AI vs env.AI_ADVISORY binding actually gets called.
+import { describe, expect, it, vi } from "vitest";
+import { runAiSlopForAdvisory } from "../../src/queue/processors";
+import { createTestEnv } from "../helpers/d1";
+import type { PullRequestFileRecord, RepositorySettings } from "../../src/types";
+import type { buildPullRequestAdvisory } from "../../src/rules/advisory";
+
+type Advisory = Awaited<ReturnType<typeof buildPullRequestAdvisory>>;
+
+function settingsFixture(advisoryAiRouting?: RepositorySettings["advisoryAiRouting"]): RepositorySettings {
+  return {
+    aiReviewByok: false,
+    aiReviewProvider: null,
+    advisoryAiRouting,
+  } as unknown as RepositorySettings;
+}
+
+describe("runAiSlopForAdvisory routes through AI_ADVISORY (#4364)", () => {
+  const frontierRun = vi.fn(async () => ({ response: JSON.stringify({ band: "clean", rationale: "ok" }) }));
+  const advisoryRun = vi.fn(async () => ({ response: JSON.stringify({ band: "clean", rationale: "ok" }) }));
+  const files: PullRequestFileRecord[] = [];
+  const advisory = { headSha: "abc123def456", findings: [] } as unknown as Advisory;
+
+  it("calls env.AI_ADVISORY, not the shared frontier env.AI, when settings.advisoryAiRouting.slop is true and the binding is configured", async () => {
+    frontierRun.mockClear();
+    advisoryRun.mockClear();
+    const env = createTestEnv({
+      AI_SUMMARIES_ENABLED: "true",
+      AI_PUBLIC_COMMENTS_ENABLED: "true",
+      AI: { run: frontierRun } as unknown as Ai,
+      AI_ADVISORY: { run: advisoryRun } as unknown as Ai,
+    });
+    await runAiSlopForAdvisory(env, {
+      settings: settingsFixture({ slop: true, e2eTestGen: false, planner: false, summaries: false }),
+      advisory,
+      repoFullName: "owner/repo",
+      pr: { number: 1, title: "t" },
+      author: "alice",
+      files,
+      deterministicBand: "clean",
+      confirmedContributor: true,
+    });
+    expect(advisoryRun).toHaveBeenCalled();
+    expect(frontierRun).not.toHaveBeenCalled();
+  });
+
+  it("calls the shared frontier env.AI (not AI_ADVISORY) when settings.advisoryAiRouting.slop is false/unset", async () => {
+    frontierRun.mockClear();
+    advisoryRun.mockClear();
+    const env = createTestEnv({
+      AI_SUMMARIES_ENABLED: "true",
+      AI_PUBLIC_COMMENTS_ENABLED: "true",
+      AI: { run: frontierRun } as unknown as Ai,
+      AI_ADVISORY: { run: advisoryRun } as unknown as Ai,
+    });
+    await runAiSlopForAdvisory(env, {
+      settings: settingsFixture(undefined),
+      advisory,
+      repoFullName: "owner/repo",
+      pr: { number: 2, title: "t" },
+      author: "alice",
+      files,
+      deterministicBand: "clean",
+      confirmedContributor: true,
+    });
+    expect(frontierRun).toHaveBeenCalled();
+    expect(advisoryRun).not.toHaveBeenCalled();
+  });
+
+  it("calls the shared frontier env.AI when slop is true but AI_ADVISORY is unconfigured (fail-safe fallback)", async () => {
+    frontierRun.mockClear();
+    advisoryRun.mockClear();
+    const env = createTestEnv({ AI_SUMMARIES_ENABLED: "true", AI_PUBLIC_COMMENTS_ENABLED: "true", AI: { run: frontierRun } as unknown as Ai });
+    await runAiSlopForAdvisory(env, {
+      settings: settingsFixture({ slop: true, e2eTestGen: false, planner: false, summaries: false }),
+      advisory,
+      repoFullName: "owner/repo",
+      pr: { number: 3, title: "t" },
+      author: "alice",
+      files,
+      deterministicBand: "clean",
+      confirmedContributor: true,
+    });
+    expect(frontierRun).toHaveBeenCalled();
+    expect(advisoryRun).not.toHaveBeenCalled();
+  });
+});
diff --git a/test/unit/advisory-ai-routing-config-engine.test.ts b/test/unit/advisory-ai-routing-config-engine.test.ts
new file mode 100644
index 0000000000..f67344d843
--- /dev/null
+++ b/test/unit/advisory-ai-routing-config-engine.test.ts
@@ -0,0 +1,46 @@
+// Mirror of the app suite pointed at the gittensory-engine copy so the extracted module owns its branch coverage (#2280).
+import { describe, expect, it } from "vitest";
+import { DEFAULT_ADVISORY_AI_ROUTING, normalizeAdvisoryAiRoutingConfig } from "../../packages/gittensory-engine/src/review/advisory-ai-routing-config";
+
+describe("normalizeAdvisoryAiRoutingConfig", () => {
+  it("returns the all-off default when input is undefined (no warnings)", () => {
+    const warnings: string[] = [];
+    expect(normalizeAdvisoryAiRoutingConfig(undefined, warnings)).toEqual(DEFAULT_ADVISORY_AI_ROUTING);
+    expect(warnings).toEqual([]);
+  });
+
+  it("normalizes a fully-valid config", () => {
+    const warnings: string[] = [];
+    expect(normalizeAdvisoryAiRoutingConfig({ slop: true, e2eTestGen: true, planner: true, summaries: true }, warnings)).toEqual({
+      slop: true,
+      e2eTestGen: true,
+      planner: true,
+      summaries: true,
+    });
+    expect(warnings).toEqual([]);
+  });
+
+  it.each(["slop", "e2eTestGen", "planner", "summaries"] as const)("defaults %s to false when omitted", (field) => {
+    const warnings: string[] = [];
+    expect(normalizeAdvisoryAiRoutingConfig({}, warnings)[field]).toBe(false);
+    expect(warnings).toEqual([]);
+  });
+
+  it.each(["slop", "e2eTestGen", "planner", "summaries"] as const)("falls back to false and warns on a non-boolean %s", (field) => {
+    const warnings: string[] = [];
+    const cfg = normalizeAdvisoryAiRoutingConfig({ [field]: "yes" }, warnings);
+    expect(cfg[field]).toBe(false);
+    expect(warnings).toEqual([`settings.advisoryAiRouting.${field} must be a boolean; using the default "false".`]);
+  });
+
+  it.each([
+    ["an array", []],
+    ["null", null],
+    ["a string", "on"],
+    ["a number", 1],
+  ])("normalizes a malformed top-level value (%s) back to the all-off default", (_label, badInput) => {
+    const warnings: string[] = [];
+    expect(normalizeAdvisoryAiRoutingConfig(badInput, warnings)).toEqual(DEFAULT_ADVISORY_AI_ROUTING);
+    expect(warnings).toEqual(["settings.advisoryAiRouting must be an object; using the default (every capability off)."]);
+  });
+});
diff --git a/test/unit/advisory-ai-routing-config.test.ts b/test/unit/advisory-ai-routing-config.test.ts
new file mode 100644
index 0000000000..2b777eb9e8
--- /dev/null
+++ b/test/unit/advisory-ai-routing-config.test.ts
@@ -0,0 +1,52 @@
+import { describe, expect, it } from "vitest";
+import { DEFAULT_ADVISORY_AI_ROUTING, normalizeAdvisoryAiRoutingConfig } from "../../src/review/advisory-ai-routing-config";
+
+describe("normalizeAdvisoryAiRoutingConfig", () => {
+  it("returns the all-off default when input is undefined (no warnings)", () => {
+    const warnings: string[] = [];
+    expect(normalizeAdvisoryAiRoutingConfig(undefined, warnings)).toEqual(DEFAULT_ADVISORY_AI_ROUTING);
+    expect(warnings).toEqual([]);
+  });
+
+  it("normalizes a fully-valid config", () => {
+    const warnings: string[] = [];
+    expect(normalizeAdvisoryAiRoutingConfig({ slop: true, e2eTestGen: true, planner: true, summaries: true }, warnings)).toEqual({
+      slop: true,
+      e2eTestGen: true,
+      planner: true,
+      summaries: true,
+    });
+    expect(warnings).toEqual([]);
+  });
+
+  it.each(["slop", "e2eTestGen", "planner", "summaries"] as const)("defaults %s to false when omitted", (field) => {
+    const warnings: string[] = [];
+    expect(normalizeAdvisoryAiRoutingConfig({}, warnings)[field]).toBe(false);
+    expect(warnings).toEqual([]);
+  });
+
+  it.each(["slop", "e2eTestGen", "planner", "summaries"] as const)("falls back to false and warns on a non-boolean %s", (field) => {
+    const warnings: string[] = [];
+    const cfg = normalizeAdvisoryAiRoutingConfig({ [field]: "yes" }, warnings);
+    expect(cfg[field]).toBe(false);
+    expect(warnings).toEqual([`settings.advisoryAiRouting.${field} must be a boolean; using the default "false".`]);
+  });
+
+  it("normalizes one valid field alongside one invalid field independently", () => {
+    const warnings: string[] = [];
+    const cfg = normalizeAdvisoryAiRoutingConfig({ slop: true, planner: "nope" }, warnings);
+    expect(cfg).toEqual({ slop: true, e2eTestGen: false, planner: false, summaries: false });
+    expect(warnings).toEqual([`settings.advisoryAiRouting.planner must be a boolean; using the default "false".`]);
+  });
+
+  it.each([
+    ["an array", []],
+    ["null", null],
+    ["a string", "on"],
+    ["a number", 1],
+  ])("normalizes a malformed top-level value (%s) back to the all-off default", (_label, badInput) => {
+    const warnings: string[] = [];
+    expect(normalizeAdvisoryAiRoutingConfig(badInput, warnings)).toEqual(DEFAULT_ADVISORY_AI_ROUTING);
+    expect(warnings).toEqual(["settings.advisoryAiRouting must be an object; using the default (every capability off)."]);
+  });
+});
diff --git a/test/unit/focus-manifest.test.ts b/test/unit/focus-manifest.test.ts
index ac972d98ae..3a5258fa38 100644
--- a/test/unit/focus-manifest.test.ts
+++ b/test/unit/focus-manifest.test.ts
@@ -350,6 +350,7 @@ describe(".gittensory.yml.example field-exhaustiveness (#1670)", () => {
     linkedIssueHardRules: "linkedIssueHardRules:",
     unlinkedIssueGuardrail: "unlinkedIssueGuardrail:",
     screenshotTableGate: "screenshotTableGate:",
+    advisoryAiRouting: "advisoryAiRouting:",
   } satisfies Record<Exclude<keyof FocusManifestSettings, (typeof SETTINGS_GATE_ALIASED_FIELDS)[number]>, string>;
 
   it.each(Object.entries(SETTINGS_FIELD_TOKENS))("documents settings.%s", (_field, token) => {
@@ -2807,6 +2808,36 @@ describe("parseFocusManifest settings override + resolveEffectiveSettings", () =
     expect(parsed.warnings).toContain(`Manifest "settings.screenshotTableGate" must be an object; ignoring it and keeping any existing policy.`);
   });
 
+  it("wires settings.advisoryAiRouting into the manifest parser as a sparse override (#4364)", () => {
+    const parsed = parseFocusManifest({ settings: { advisoryAiRouting: { slop: true, summaries: true } } });
+    expect(parsed.settings.advisoryAiRouting).toEqual({ slop: true, summaries: true });
+    expect(parsed.warnings).toEqual([]);
+  });
+
+  it("resolveEffectiveSettings merges a partial advisoryAiRouting override without clearing the lower-layer fields (#4364)", () => {
+    const db = { advisoryAiRouting: { slop: false, e2eTestGen: true, planner: false, summaries: true } } as unknown as RepositorySettings;
+    const eff = resolveEffectiveSettings(db, parseFocusManifest({ settings: { advisoryAiRouting: { slop: true } } }));
+    expect(eff.advisoryAiRouting).toEqual({ slop: true, e2eTestGen: true, planner: false, summaries: true });
+  });
+
+  it("resolveEffectiveSettings falls back to the all-off built-in default when the DB layer has no advisoryAiRouting at all (#4364)", () => {
+    const db = {} as unknown as RepositorySettings;
+    const eff = resolveEffectiveSettings(db, parseFocusManifest({ settings: { advisoryAiRouting: { planner: true } } }));
+    expect(eff.advisoryAiRouting).toEqual({ slop: false, e2eTestGen: false, planner: true, summaries: false });
+  });
+
+  it("drops a malformed advisoryAiRouting.slop field instead of replacing existing policy with defaults (#4364)", () => {
+    const parsed = parseFocusManifest({ settings: { advisoryAiRouting: { slop: "yes", planner: true } } });
+    expect(parsed.settings.advisoryAiRouting).toEqual({ planner: true });
+    expect(parsed.warnings.some((w) => w.includes("settings.advisoryAiRouting.slop"))).toBe(true);
+  });
+
+  it("warns and ignores a malformed top-level advisoryAiRouting value (#4364)", () => {
+    const parsed = parseFocusManifest({ settings: { advisoryAiRouting: "oops" } });
+    expect(parsed.settings.advisoryAiRouting).toBeUndefined();
+    expect(parsed.warnings).toContain(`Manifest "settings.advisoryAiRouting" must be an object; ignoring it and keeping any existing policy.`);
+  });
+
   it("parses aiReview from settings: and lets gate.aiReview win in resolveEffectiveSettings", () => {
     const parsed = parseFocusManifest({ settings: { aiReviewMode: "advisory", aiReviewByok: true } });
     expect(parsed.settings.aiReviewMode).toBe("advisory");
diff --git a/test/unit/selfhost-ai.test.ts b/test/unit/selfhost-ai.test.ts
index 6cffa8b810..50820780b3 100644
--- a/test/unit/selfhost-ai.test.ts
+++ b/test/unit/selfhost-ai.test.ts
@@ -2,7 +2,7 @@ import { chmodSync, mkdirSync, mkdtempSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { delimiter, join } from "node:path";
 import { afterEach, describe, expect, it, vi } from "vitest";
-import { assertNoLegacySharedAiEnv, buildProvider, claudeErrorStatus, codexErrorFromStdout, createAnthropicAi, createChainAi, createClaudeCodeAi, createCodexAi, createOpenAiCompatibleAi, createSelfHostAi, extractCliText, extractCliUsage, isAiProviderHealthy, markAiProviderUnhealthyAtBoot, resetAiProviderCircuitBreakerForTest, resetAiProviderHealthForTest, resolveAiReviewerPlan, resolveClaudeCliTimeoutMs, resolveCodexAuthPath, resolveCodexCliTimeoutMs, resolveCodexEffort, resolveCodexFirstOutputTimeoutMs, resolveEffort, resolveModel, resolveProviderNames, resolveRequiredCliProviders, resolveSubscriptionCliPath, redactSecrets, routeProviders, shouldMarkAiProviderUnhealthyAtBoot, subscriptionCliEnv } from "../../src/selfhost/ai";
+import { assertNoLegacySharedAiEnv, buildProvider, claudeErrorStatus, codexErrorFromStdout, createAnthropicAi, createChainAi, createClaudeCodeAi, createCodexAi, createOpenAiCompatibleAi, createSelfHostAi, extractCliText, extractCliUsage, isAiProviderHealthy, markAiProviderUnhealthyAtBoot, resetAiProviderCircuitBreakerForTest, resetAiProviderHealthForTest, resolveAiReviewerPlan, resolveClaudeCliTimeoutMs, resolveCodexAuthPath, resolveCodexCliTimeoutMs, resolveCodexEffort, resolveCodexFirstOutputTimeoutMs, resolveEffort, resolveModel, resolveProviderNames, resolveRequiredCliProviders, resolveSubscriptionCliPath, redactSecrets, routeProviders, shouldMarkAiProviderUnhealthyAtBoot, subscriptionCliEnv, withAdvisoryAiEnv } from "../../src/selfhost/ai";
 import { labelSelfHostReviewerModel, labelSelfHostReviewerModels } from "../../src/selfhost/ai-config";
 import { renderMetrics, resetMetrics } from "../../src/selfhost/metrics";
 
@@ -1778,3 +1778,30 @@ describe("redactSecrets — strip credentials from untrusted CLI stderr before i
     expect(redactSecrets("disk-usage-report-2024-summary failed")).toBe("disk-usage-report-2024-summary failed");
   });
 });
+
+describe("withAdvisoryAiEnv (#4364 — per-capability local-inference routing)", () => {
+  const frontierAi = { run: async () => ({ response: "frontier" }) };
+  const advisoryAi = { run: async () => ({ response: "advisory" }) };
+
+  it("swaps .AI to AI_ADVISORY when the capability opted in AND the binding is configured", () => {
+    const env = { AI: frontierAi, AI_ADVISORY: advisoryAi } as unknown as Env;
+    const result = withAdvisoryAiEnv(env, true);
+    expect(result.AI).toBe(advisoryAi);
+    // every other field is untouched (shallow spread, not a new unrelated object)
+    expect(result.AI_ADVISORY).toBe(advisoryAi);
+  });
+
+  it("leaves .AI unchanged when the capability did not opt in, even though the binding is configured", () => {
+    const env = { AI: frontierAi, AI_ADVISORY: advisoryAi } as unknown as Env;
+    const result = withAdvisoryAiEnv(env, false);
+    expect(result).toBe(env); // same reference — no-op, not just an equal value
+    expect(result.AI).toBe(frontierAi);
+  });
+
+  it("falls back to the shared frontier chain when opted in but AI_ADVISORY is unconfigured (byte-identical fail-safe)", () => {
+    const env = { AI: frontierAi } as unknown as Env;
+    const result = withAdvisoryAiEnv(env, true);
+    expect(result).toBe(env);
+    expect(result.AI).toBe(frontierAi);
+  });
+});