feat(localai): Upgrade to v3.9.0 with Agent Jobs and Memory Reclaimer

Upgrade LocalAI from v2.25.0 to v3.9.0 with new features:

- **Agent Jobs Panel**: Schedule and manage background agentic tasks
- **Memory Reclaimer**: LRU eviction for loaded models, automatic VRAM cleanup
- **VibeVoice backend**: New voice synthesis support

Update README with:
- v3.9 feature highlights
- Complete CLI command reference
- Model presets table (tinyllama, phi2, mistral, gte-small)
- API endpoints documentation
- SecuBox Couche 2 integration notes

This is part of the v0.18 AI Gateway roadmap.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

package/secubox/secubox-app-localai/Makefile

@@ -9,7 +9,7 @@
 include $(TOPDIR)/rules.mk
 
 PKG_NAME:=secubox-app-localai
-PKG_VERSION:=2.25.0
+PKG_VERSION:=3.9.0
 PKG_RELEASE:=1
 
 PKG_LICENSE:=MIT

package/secubox/secubox-app-localai/README.md

@@ -2,10 +2,33 @@
 
 Native LLM server with OpenAI-compatible REST API. Supports GGUF models on ARM64 and x86_64.
 
+**Version**: 3.9.0
+
+## Features
+
+- OpenAI-compatible REST API (`/v1/chat/completions`, `/v1/completions`, `/v1/embeddings`)
+- GGUF model support (LLaMA, Mistral, Phi, TinyLlama, etc.)
+- ARM64 and x86_64 architectures
+- **Agent Jobs Panel** (v3.9) — Schedule and manage background agentic tasks
+- **Memory Reclaimer** (v3.9) — LRU eviction for loaded models
+- Embeddings support (GTE-Small preset)
+
 ## Installation
 
 ```sh
 opkg install secubox-app-localai
+
+# Download the binary (first run)
+localaictl install
+
+# Install a model
+localaictl model-install tinyllama
+
+# Enable and start service
+uci set localai.main.enabled=1
+uci commit localai
+/etc/init.d/localai enable
+/etc/init.d/localai start
 ```
 
 ## Configuration
@@ -15,40 +38,68 @@ UCI config file: `/etc/config/localai`
 ```
 config localai 'main'
     option enabled '0'
-    option port '8080'
+    option api_port '8081'
+    option api_host '0.0.0.0'
     option models_path '/srv/localai/models'
+    option threads '4'
+    option context_size '2048'
 ```
 
-## Usage
+## Model Presets
+
+| Preset | Size | Description |
+|--------|------|-------------|
+| tinyllama | 669MB | TinyLlama 1.1B (chat) |
+| phi2 | 1.6GB | Microsoft Phi-2 (chat) |
+| mistral | 4.1GB | Mistral 7B Instruct (chat) |
+| gte-small | 67MB | GTE Small (embeddings) |
 
 ```sh
-# Install the binary (downloaded on first run)
+localaictl model-install tinyllama
-localaictl install
+localaictl model-install gte-small   # For embeddings
-
-# Start / stop the service
-localaictl start
-localaictl stop
-
-# Check status
-localaictl status
-
-# Download a model
-localaictl model-pull <model-name>
 ```
 
-The binary is downloaded from GitHub releases on first `localaictl install`.
+## CLI Commands
 
-## Features
+```sh
+# Install/Uninstall
+localaictl install          # Download binary from GitHub
+localaictl uninstall        # Remove binary
 
-- OpenAI-compatible REST API
+# Service
-- GGUF model support (LLaMA, Mistral, Phi, TinyLlama, etc.)
+localaictl start            # Start service
-- ARM64 and x86_64 architectures
+localaictl stop             # Stop service
+localaictl restart          # Restart service
+localaictl status           # Show status
+localaictl logs [-f]        # Show logs
+
+# Models
+localaictl models           # List installed models
+localaictl model-install <name>  # Install model
+localaictl model-remove <name>   # Remove model
+
+# Backends
+localaictl backends         # List available backends
+```
+
+## API Endpoints
+
+Default port: 8081
+
+| Endpoint | Description |
+|----------|-------------|
+| `/v1/models` | List loaded models |
+| `/v1/chat/completions` | Chat completion |
+| `/v1/completions` | Text completion |
+| `/v1/embeddings` | Generate embeddings |
+| `/readyz` | Health check |
 
 ## Files
 
-- `/etc/config/localai` -- UCI configuration
+- `/etc/config/localai` — UCI configuration
-- `/usr/sbin/localaictl` -- controller CLI
+- `/usr/sbin/localaictl` — Controller CLI
-- `/srv/localai/models/` -- model storage directory
+- `/usr/bin/local-ai` — Binary (downloaded)
+- `/srv/localai/models/` — Model storage
 
 ## Dependencies
 
@@ -57,6 +108,18 @@ The binary is downloaded from GitHub releases on first `localaictl install`.
 - `wget-ssl`
 - `ca-certificates`
 
+## SecuBox Integration
+
+LocalAI serves as the inference backend for SecuBox AI Gateway (Couche 2):
+- Local-first LLM inference
+- MCP Server integration for agent tools
+- Supports Threat Analyst, CVE Triage, and other autonomous agents
+
 ## License
 
 MIT
+
+## Sources
+
+- [LocalAI GitHub](https://github.com/mudler/LocalAI)
+- [LocalAI v3.9.0 Release](https://github.com/mudler/LocalAI/releases)

package/secubox/secubox-app-localai/files/usr/sbin/localaictl

@@ -8,7 +8,7 @@ CONFIG="localai"
 BINARY="/usr/bin/local-ai"
 DATA_DIR="/srv/localai"
 BACKEND_ASSETS="/usr/share/localai/backend-assets"
-LOCALAI_VERSION="2.25.0"
+LOCALAI_VERSION="3.9.0"
 
 usage() {
 	cat <<'EOF'