feat(inferno): move Potato OS to the standalone inferno package

AGENTS.md

@@ -6,7 +6,7 @@ Core application code lives in `core/`:
 - `core/main.py`: FastAPI entrypoint, API routes, orchestrator loop, lifespan.
 - `core/model_state.py`: Model registry, settings persistence, projector helpers.
 - `core/runtime_state.py`: Runtime config, dual-runtime slot discovery, system metrics, power calibration.
-- `core/inferno/`: Inference layer — backend proxy, model family classification, LiteRT adapter, launch config builder. See `core/inferno/__init__.py` for boundary contract.
+- `inferno` (external package, [`potato-os/inferno`](https://github.com/potato-os/inferno)): Inference layer — backend proxy, model family classification, LiteRT adapter, launch config builder, model registry, runtime management, orchestration. Installed via `requirements.txt`.
 - `core/assets/`: Frontend — `index.html`, `shell.css`, `shell.js` (platform shell), vendor libs.
 - `core/rig_envelope.py`: RIG step envelope validation (MS/TS contract checks).
 
@@ -66,6 +66,8 @@ These commands are written for a **macOS dev environment**. `COPYFILE_DISABLE=1`
     --rsync-path="SUDO_ASKPASS=/opt/potato/bin/askpass.sh sudo -A rsync" \
     -e "ssh -o StrictHostKeyChecking=accept-new" \
     core/ pi@potato.local:/opt/potato/core/
+  sshpass -e ssh pi@potato.local \
+    "echo raspberry | sudo -S /opt/potato/venv/bin/pip install -r /opt/potato/core/requirements.txt"
   ```
 - Fast apps deploy:
   ```

apps/chat/routes.py

@@ -9,16 +9,16 @@
 from fastapi import APIRouter, Depends, HTTPException, Request, Response
 from fastapi.responses import JSONResponse, StreamingResponse
 
+from inferno import BackendProxyError, ChatRepositoryManager
+
 try:
     from core.deps import get_runtime, get_chat_repository
     from core.model_state import apply_model_chat_defaults
-    from core.inferno import BackendProxyError, ChatRepositoryManager
     from core.runtime_state import RuntimeConfig
     from core.settings import merge_active_model_chat_defaults, merge_chat_defaults
 except ModuleNotFoundError:
     from deps import get_runtime, get_chat_repository  # type: ignore[no-redef]
     from model_state import apply_model_chat_defaults  # type: ignore[no-redef]
-    from inferno import BackendProxyError, ChatRepositoryManager  # type: ignore[no-redef]
     from runtime_state import RuntimeConfig  # type: ignore[no-redef]
     from settings import merge_active_model_chat_defaults, merge_chat_defaults  # type: ignore[no-redef]
 

bin/install_dev.sh

@@ -301,8 +301,8 @@ rm -f "${sudoers_terminal_tmp}"
 
 sudoers_ota_tmp="$(mktemp)"
 cat > "${sudoers_ota_tmp}" <<'SUDOERS'
-potato ALL=(root) NOPASSWD: /bin/chown -R potato\:potato /opt/potato/app
-potato ALL=(root) NOPASSWD: /usr/bin/chown -R potato\:potato /opt/potato/app
+potato ALL=(root) NOPASSWD: /bin/chown -R potato\:potato /opt/potato/core
+potato ALL=(root) NOPASSWD: /usr/bin/chown -R potato\:potato /opt/potato/core
 potato ALL=(root) NOPASSWD: /bin/chown -R potato\:potato /opt/potato/bin
 potato ALL=(root) NOPASSWD: /usr/bin/chown -R potato\:potato /opt/potato/bin
 SUDOERS

bin/start_litert.sh

@@ -29,7 +29,7 @@ export POTATO_BASE_DIR
 cd "${POTATO_BASE_DIR}"
 
 exec "${POTATO_VENV_DIR}/bin/uvicorn" \
-  core.inferno.litert_adapter:app \
+  inferno.litert_adapter:app \
   --host 0.0.0.0 \
   --port "${POTATO_LLAMA_PORT}" \
   --workers 1

bin/start_llama.sh

@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
 # Thin wrapper around llama-server — all business logic lives in Python
-# (core.inferno.launch_config.build_llama_server_args).
+# (inferno.launch_config.build_llama_server_args).
 #
 # This script only handles:
 #   1. Dynamic library path for GGML backends

core/deps.py

@@ -4,12 +4,12 @@
 
 from fastapi import Request
 
+from inferno import ChatRepositoryManager
+
 try:
     from core.runtime_state import RuntimeConfig
-    from core.inferno import ChatRepositoryManager
 except ModuleNotFoundError:
     from runtime_state import RuntimeConfig  # type: ignore[no-redef]
-    from inferno import ChatRepositoryManager  # type: ignore[no-redef]
 
 
 def get_runtime(request: Request) -> RuntimeConfig: