Frequently Asked Questions

• OS: Windows 10 21H2 (64-bit) or Windows 11
• RAM: 4 GB minimum, 8 GB recommended (for the AI module)
• Disk: 500 MB for the app + ~4 GB for GGUF model if you use AI
• Permissions: Administrator account — Coclico requests UAC elevation at startup
• GPU (optional): Vulkan-compatible card to accelerate local AI inference

The .NET 10 runtime is bundled in the installer — nothing else to install.

Yes, this is normal for software whose code signature is not yet recognized by Microsoft. Coclico is open-source and you can inspect its source code on GitHub. To proceed:

1. Click "More info" in the SmartScreen window
2. Click "Run anyway"

Alternatively, compile Coclico directly from source using dotnet build after cloning the repository.

No. Coclico requires administrator rights to access low-level Windows APIs: WMI, kernel memory (ntdll), system registry, and process management. UAC elevation is requested once at startup. A reduced-functionality "user-mode" version is planned for a future release.

No. The RAM Cleaner uses official Windows APIs to trim process working sets and flush standby lists — the same operation Windows performs internally under memory pressure. Running applications continue normally. Trimmed memory pages are paged to disk and reloaded on demand. The visible effect is an immediate reduction in used RAM with a slight temporary increase in disk I/O if applications are currently idle.

Coclico only deletes known-safe file categories (temp, browser cache, logs, recycle bin, prefetch, etc.). No personal documents or critical system files are targeted. Each category can be individually enabled/disabled before running. A detailed report lists files to be deleted before any action is taken.

Go to Settings → Appearance. The default accent color is #6366F1 (Indigo). Enter any hex value or choose from presets. Changes are applied immediately without restart via ThemeService which dynamically updates WPF resources.

No, absolutely not. Coclico enforces a Zero Cloud principle: all AI inference runs locally on your machine via LLamaSharp 0.26 with a GGUF model. No network request is made for AI. Your system data, conversations and Flow Chains never leave your machine.

Coclico works with quantized GGUF models (Q4_K_M or Q5_K_M). Recommended models:

• Llama 3.2 3B Q5_K_M — lightweight, fast on CPU, good quality for system tasks
• Mistral 7B Q4_K_M — better quality, requires 8 GB RAM
• Phi-3 Mini 3.8B Q5 — excellent performance/quality ratio on CPU

Place the file in %APPDATA%\Coclico\resource\model\ and name it IA-support-chat.gguf. Vulkan GPU acceleration is automatic if supported.

The AutoPatcher is Coclico's autonomous system that can propose code patches to improve itself (Phase 3.4). In Audit Only mode (default), AI-generated patches are logged in the audit log but never applied automatically. You must manually approve them via ApproveAndApplyAsync(). Each patch is validated by the Digital Twin (cyclomatic complexity check) and a rollback snapshot is created before any disk write.

There is no hard limit in the code. In practice, chains over 100 nodes become difficult to maintain. Coclico provides 28 node types (NodeType enum). For complex automations, create multiple modular chains and link them via inter-chain calls.

Behavior depends on the OnErrorAction property configured on each node:

• ContinueNext (default) — error is logged, execution continues with the next node
• StopChain — execution stops immediately, no subsequent nodes run
• SkipToEnd — all intermediate nodes are skipped, only the last chain node executes

Yes, via SetRegistryValue and DeleteRegistryKey nodes. However, each operation passes through WorkflowPipelineExecutionService which applies ISecurityPolicy before execution. By default, modifications to HKLM\SYSTEM\CurrentControlSet and other sensitive keys are blocked. Customize the policy via %APPDATA%\Coclico\security-policy.json.

No. In idle mode (system tray icon), Coclico consumes less than 20 MB RAM and under 0.1% CPU. The telemetry service (DynamicTracerService) samples the system once per second with negligible overhead. The OptimizationEngine runs every 30 seconds using ArrayPool buffers to avoid GC allocations.

No. Coclico uses two fully isolated LLM contexts (dual-executor): one for user chat, one for the background OptimizationEngine. The UI thread never blocks waiting for an LLM response — results are returned via async/await and UI updates use Dispatcher.InvokeAsync. Inference may take 5 to 30 seconds depending on model and hardware, without ever freezing the UI.

Several options in Settings → General:

• Disable automatic Dashboard refresh (change from 1s to 5s)
• Reduce OptimizationEngine frequency (60s instead of 30s)
• Disable the AI module if not needed (unloads the GGUF model from memory)
• Disable toast notifications to reduce UI redraws