fix(uninstall): delete all installed host models

docs/inference/choose-local-inference-server.mdx

@@ -153,8 +153,10 @@ Refer to [Host Files and State](../../reference/host-files-and-state) before del
 
 When you intend to remove the entire NemoClaw installation, run `$$nemoclaw uninstall`.
 Full uninstall verifies exact managed container ownership before it removes the runtime and its state.
-It preserves the shared Hugging Face cache used by vLLM.
-Add `--delete-models` only when you also accept deletion of NemoClaw-pulled Ollama models.
+It preserves the shared Hugging Face cache used by vLLM by default.
+Add `--delete-models` only when you accept deletion of every model reported by the host's local Ollama inventory and all non-credential data in the current user's shared Hugging Face cache.
+This opt-in can delete cached files that other applications installed or use.
+It preserves the Hugging Face `token` and `stored_tokens` authentication files.
 If cleanup reports an ownership or Docker error, leave the state files in place, resolve the reported resource, and rerun uninstall.
 After a successful full uninstall, `docker container inspect nemoclaw-vllm` must report no object.
 

docs/inference/set-up-llama-cpp.mdx

@@ -246,7 +246,10 @@ It preserves `~/.cache/huggingface/` because other applications can use that cac
 If exact cleanup fails, NemoClaw preserves the sandbox registry entry and ownership state for a retry.
 
 For full NemoClaw removal, run `$$nemoclaw uninstall`.
-Full uninstall applies the same exact-ownership checks and preserves the shared Hugging Face cache.
+Full uninstall applies the same exact-ownership checks and preserves the shared Hugging Face cache by default.
+Pass `--delete-models` only when you accept deletion of all non-credential data in the current user's shared cache after managed model runtimes stop.
+The same flag also deletes every model installed in the host's local Ollama inventory.
+Uninstall preserves the Hugging Face `token` and `stored_tokens` authentication files.
 
 ## Related Topics
 

docs/manage-sandboxes/uninstall-nemoclaw.mdx

@@ -24,7 +24,7 @@ It preserves resources from a separate OpenClaw installation while continuing to
 |---|---|
 | `--yes` | Skip the confirmation prompt. |
 | `--keep-openshell` | Leave OpenShell binaries, NemoClaw-managed gateway service files, and local gateway state in place, and do not stop the host gateway process. |
-| `--delete-models` | Also remove NemoClaw-pulled Ollama models. |
+| `--delete-models` | Delete every model reported by the host's local Ollama inventory and all non-credential data in the current user's shared `~/.cache/huggingface/` cache after managed model runtimes stop. Hugging Face authentication files remain. |
 | `--destroy-user-data` | Also remove preserved user data in `rebuild-backups/`, `backups/`, and `sandboxes.json`. |
 | `--all-gateway-ports` | Uninstall every gateway port on the host instead of only the selected one. |
 | `--gateway <name>` | Optional consistency check; must match the name derived from `NEMOCLAW_GATEWAY_PORT`. |
@@ -33,6 +33,35 @@ It preserves resources from a separate OpenClaw installation while continuing to
 For example, `NEMOCLAW_GATEWAY_PORT=9123 $$nemoclaw uninstall` selects `nemoclaw-9123` and its port-scoped state.
 Do not use `--gateway` to select another instance; when supplied for compatibility, its value must match the derived name or uninstall stops before cleanup.
 
+## Delete Host Models
+
+The `--delete-models` flag is an explicit opt-in for deleting host model files.
+
+<Warning>
+This flag deletes every model that the local Ollama inventory reports, including models installed or used by other applications.
+It also deletes all non-credential data in the current user's `~/.cache/huggingface/` directory, which other applications can use for models, datasets, and other cached assets.
+Affected applications must download and verify their cached files again.
+The Hugging Face `token` and `stored_tokens` authentication files remain, so this cleanup does not sign the current user out.
+</Warning>
+
+Run full uninstall with model deletion:
+
+```bash
+$$nemoclaw uninstall --delete-models
+```
+
+NemoClaw stops and verifies its managed local and distributed model runtimes before it deletes non-credential data from the local Hugging Face cache.
+It validates that `~/.cache/` and `~/.cache/huggingface/` are real, current-user-owned directories that are not symlinks or group- or world-writable.
+It inventories Ollama before it deletes any Ollama model.
+An Ollama inventory error, model deletion error, unsafe cache path, or cache-data deletion error makes uninstall exit nonzero.
+Cleanup can partially complete before an error.
+Resolve the reported error, inspect the remaining models and runtimes, and rerun uninstall.
+
+Without `--delete-models`, uninstall preserves both model stores.
+When sibling gateway environments remain, uninstall preserves both model stores even if you pass `--delete-models`.
+An all-gateway-port sweep deletes them only after every selected gateway cleanup succeeds and no sibling environment remains.
+The flag does not scan arbitrary directories or delete model caches on remote peers.
+
 ## Uninstall Every Gateway Port
 
 One uninstall removes one gateway port.

docs/reference/commands.mdx

@@ -3954,8 +3954,15 @@ These host-local checks run before NemoClaw deletes their state.
 If Docker is unavailable or a resource does not match its persisted ownership state, uninstall exits nonzero before the remaining uninstall steps and preserves that state for recovery.
 Host-local cleanup can partially complete before an error.
 Restore Docker access or resolve the named ownership conflict, inspect the remaining container and network, and retry uninstall.
-Managed llama.cpp and vLLM model files remain in the shared Hugging Face cache.
-The `--delete-models` flag does not remove this shared cache.
+Managed llama.cpp and vLLM model files remain in the shared Hugging Face cache by default.
+With `--delete-models`, uninstall deletes every model in the local Ollama inventory and all non-credential data in the current user's shared `~/.cache/huggingface/` cache.
+This opt-in can delete cached files that other applications installed or use.
+It preserves the Hugging Face `token` and `stored_tokens` authentication files.
+NemoClaw stops and verifies its managed local and distributed model runtimes before it deletes non-credential data from the local Hugging Face cache.
+It does not scan arbitrary directories or delete model caches on remote peers.
+When sibling gateway environments remain, uninstall preserves both model stores even if you pass `--delete-models`.
+An Ollama inventory error, model deletion error, unsafe cache path, or cache-data deletion error makes uninstall exit nonzero.
+Cleanup can partially complete before an error, so resolve the reported error and rerun uninstall.
 It does not use the legacy `gateway destroy` command for that gateway.
 <AgentOnly variant="openclaw">
 
@@ -3981,7 +3988,7 @@ Uninstall does not otherwise target the declared external directory.
 |---|---|
 | `--yes` | Skip the confirmation prompt |
 | `--keep-openshell` | Leave OpenShell binaries, NemoClaw-managed gateway service files, and local gateway state in place, and do not stop the host gateway process |
-| `--delete-models` | Also remove NemoClaw-pulled Ollama models. The shared Hugging Face cache remains in place. |
+| `--delete-models` | Delete every model reported by the host's local Ollama inventory and all non-credential data in the current user's shared `~/.cache/huggingface/` cache after managed model runtimes stop. Hugging Face authentication files remain. |
 | `--destroy-user-data` | Also remove preserved user data (`rebuild-backups/`, `backups/`, `sandboxes.json`) |
 | `--all-gateway-ports` | Uninstall every gateway port on the host, not only the port `NEMOCLAW_GATEWAY_PORT` selects |
 | `--gateway <name>` | Optional consistency check; must match the name derived from `NEMOCLAW_GATEWAY_PORT` |

docs/reference/host-files-and-state.mdx

@@ -93,7 +93,7 @@ Run `$$nemoclaw onboard` to register a sandbox again.
 | `~/.nemoclaw/backups/` | Workspace backups written by legacy backup helpers and some recovery flows. | Only after confirming you no longer need those workspace archives. |
 | `~/.nemoclaw/mounts/` | Default local mount points created by share or mount commands. | Unmount first, then remove unused directories. |
 | `~/.nemoclaw/blueprints/` | Cached blueprint inputs used by onboarding and sandbox recreation. | Avoid manual deletion unless you plan to rerun onboarding from fresh inputs. |
-| `~/.cache/huggingface/` | Shared Hugging Face cache used by managed vLLM and managed llama.cpp model acquisition. The llama.cpp path resolves the exact YAML-declared revision and GGUF file, then verifies its size, SHA-256 digest, and filesystem identity before launch. Other applications can use the same cache. | Only after stopping every process that uses the cache. Destroy and full uninstall preserve this shared cache, including when you pass `--delete-models`. Deleting it makes affected applications download and verify model files again. |
+| `~/.cache/huggingface/` | Shared Hugging Face cache used by managed vLLM and managed llama.cpp model acquisition. The llama.cpp path resolves the exact YAML-declared revision and GGUF file, then verifies its size, SHA-256 digest, and filesystem identity before launch. Other applications can use the same cache. | Only after stopping every process that uses the cache. Sandbox destroy and full uninstall without `--delete-models` preserve it. Full uninstall with `--delete-models` deletes its non-credential data after managed model runtimes stop and only when no sibling gateway remains. The `token` and `stored_tokens` authentication files remain. Affected applications must download and verify their cached files again. |
 
 ## Migration Snapshot Retention
 
@@ -137,8 +137,12 @@ Before full uninstall deletes shared state, it removes only an exact authenticat
 Catalog-selected host-local vLLM also requires its exact owner-only runtime receipt; legacy authenticated containers without catalog profile labels retain their existing cleanup behavior.
 It refuses the remaining uninstall steps when Docker is unavailable or persisted ownership cannot be proved.
 Resolve the reported Docker or ownership error and rerun uninstall with the state still present.
-The shared Hugging Face cache used by managed vLLM and managed llama.cpp remains in place.
-The `--delete-models` flag removes NemoClaw-pulled Ollama models, not the shared Hugging Face cache.
+The shared Hugging Face cache used by managed vLLM and managed llama.cpp remains in place by default.
+With `--delete-models`, full uninstall deletes every model in the local Ollama inventory and all non-credential data in the current user's shared Hugging Face cache.
+The flag can delete cached files that other applications installed or use.
+It preserves the Hugging Face `token` and `stored_tokens` authentication files.
+NemoClaw deletes the non-credential cache data only after managed model runtimes stop and only when no sibling gateway remains.
+An Ollama inventory error, model deletion error, unsafe cache path, or cache-data deletion error makes uninstall exit nonzero.
 
 For operational uninstall steps, refer to:
 

scripts/smoke-macos-install.sh

@@ -59,7 +59,7 @@ Options:
   --allow-existing-state      Allow running even if NemoClaw/OpenShell state already exists
   --keep-logs                 Preserve log files after success
   --remove-openshell          Allow uninstall.sh to remove openshell
-  --delete-models             Allow uninstall.sh to delete Ollama models
+  --delete-models             Allow uninstall.sh to delete all Ollama models and non-credential Hugging Face cache data (authentication files remain)
   -h, --help                  Show this help
 
 Environment:

src/commands/internal/uninstall/plan.ts

@@ -2,12 +2,11 @@
 // SPDX-License-Identifier: Apache-2.0
 
 import { Flags } from "@oclif/core";
+import { buildHostUninstallPlan } from "../../../lib/actions/uninstall/plan";
 import { CLI_DISPLAY_NAME, CLI_NAME } from "../../../lib/cli/branding";
 import { jsonFlag } from "../../../lib/cli/common-flags";
 import { NemoClawCommand } from "../../../lib/cli/nemoclaw-oclif-command";
 
-import { buildHostUninstallPlan } from "../../../lib/actions/uninstall/plan";
-
 export default class InternalUninstallPlanCommand extends NemoClawCommand {
   static hidden = true;
   static strict = true;
@@ -21,7 +20,8 @@ export default class InternalUninstallPlanCommand extends NemoClawCommand {
       description: "Accepted for parity with run-plan; ignored while planning",
     }),
     "delete-models": Flags.boolean({
-      description: `Plan removal of ${CLI_DISPLAY_NAME}-pulled Ollama models`,
+      description:
+        "Plan removal of all Ollama models and non-credential Hugging Face cache data (authentication files remain)",
     }),
     "keep-openshell": Flags.boolean({ description: "Keep the openshell binary installed" }),
     gateway: Flags.string({ description: "Gateway name", default: "nemoclaw" }),

src/commands/internal/uninstall/run-plan.ts

@@ -30,7 +30,8 @@ export default class InternalUninstallRunPlanCommand extends NemoClawCommand {
     "all-gateway-ports-child": Flags.boolean({ hidden: true }),
     "keep-openshell": Flags.boolean({ description: "Leave the openshell binary installed" }),
     "delete-models": Flags.boolean({
-      description: `Remove ${CLI_DISPLAY_NAME}-pulled Ollama models`,
+      description:
+        "Remove all Ollama models and non-credential Hugging Face cache data (authentication files remain)",
     }),
     "destroy-user-data": Flags.boolean({
       description:

src/lib/actions/root-help.ts

@@ -81,7 +81,9 @@ export function help(): void {
   lines.push(`  ${G}Uninstall flags:${R}`);
   lines.push(`    --yes${" ".repeat(29)}Skip the confirmation prompt`);
   lines.push(`    --keep-openshell${" ".repeat(18)}Leave the openshell binary installed`);
-  lines.push(`    --delete-models${" ".repeat(19)}Remove ${CLI_DISPLAY_NAME}-pulled Ollama models`);
+  lines.push(
+    `    --delete-models${" ".repeat(19)}Remove all Ollama models and non-credential Hugging Face cache data (authentication files remain)`,
+  );
   lines.push(
     `    --destroy-user-data${" ".repeat(15)}Remove preserved ~/.nemoclaw/ user data (rebuild-backups/, backups/, sandboxes.json)`,
   );

src/lib/actions/uninstall/run-plan-dual-station.test.ts

@@ -7,6 +7,8 @@ import path from "node:path";
 
 import { describe, expect, it, vi } from "vitest";
 
+import { DUAL_STATION_VLLM_RUNTIME_RECEIPT_FILE } from "../../inference/serving/managed-runtime-receipts";
+
 import {
   type RunResult,
   runUninstallPlan as runUninstallPlanBase,
@@ -90,6 +92,47 @@ describe("managed distributed vLLM runtime uninstall", () => {
     }
   });
 
+  it("stops a distributed runtime before requesting shared Hugging Face cache-data cleanup", () => {
+    const home = fs.mkdtempSync(path.join(os.tmpdir(), "nemoclaw-uninstall-cache-order-"));
+    const stateDir = path.join(home, ".nemoclaw");
+    const receiptPath = path.join(stateDir, DUAL_STATION_VLLM_RUNTIME_RECEIPT_FILE);
+    const cacheDir = path.join(home, ".cache", "huggingface");
+    fs.mkdirSync(stateDir, { mode: 0o700 });
+    fs.writeFileSync(receiptPath, "{}\n", { mode: 0o600 });
+    fs.mkdirSync(`${receiptPath}.ssh-binding`, { mode: 0o700 });
+    fs.mkdirSync(cacheDir, { recursive: true });
+    const runDualStationRuntimeCleanup = vi.fn(() => ok());
+    const runHuggingFaceCacheDataCleanup = vi.fn(() => ok());
+
+    try {
+      const result = runUninstallPlan(
+        { assumeYes: true, deleteModels: true, keepOpenShell: true },
+        {
+          commandExists: (command) => command === "openshell",
+          env: { HOME: home, TMPDIR: home } as NodeJS.ProcessEnv,
+          existsSync: fs.existsSync,
+          isTty: false,
+          log: vi.fn(),
+          rmSync: vi.fn(),
+          run: okWithKnownGatewayList,
+          runDualStationRuntimeCleanup,
+          runHuggingFaceCacheDataCleanup,
+        },
+      );
+
+      expect(result.exitCode).toBe(0);
+      expect(runDualStationRuntimeCleanup).toHaveBeenCalledOnce();
+      expect(runHuggingFaceCacheDataCleanup).toHaveBeenCalledWith(
+        expect.objectContaining({ stdio: "inherit" }),
+      );
+      expect(runDualStationRuntimeCleanup.mock.invocationCallOrder[0]).toBeLessThan(
+        runHuggingFaceCacheDataCleanup.mock.invocationCallOrder[0],
+      );
+    } finally {
+      fs.rmSync(home, { recursive: true, force: true });
+    }
+  });
+
   it("finds host-global pair ownership when the final gateway uses a non-default port", async () => {
     const home = fs.mkdtempSync(path.join(os.tmpdir(), "nemoclaw-uninstall-dual-port-"));
     const port = 9123;