FlowyAIPC
  • English

    • Advanced Operations

    Switch Conversation Target (Switch Agent)

    1. Click the @ button next to the input box
    2. Select the target agent

    Snipaste_2026-06-01_12-29-33

    Snipaste_2026-06-01_12-30-42

    Snipaste_2026-06-01_12-57-36

    Enable Security Protection

    Effect: Prevents AI from accidentally deleting/modifying critical files

    1. Click Settings → Security Protection

    Snipaste_2026-06-01_14-14-47

    1. Add important folders to the My Protected Directories list

    Snipaste_2026-06-01_14-15-11

    Snipaste_2026-06-01_14-15-40

    Enable Local Model Lite Mode

    NOTE

    1. What is this feature for?

    Local Model Lite Mode is a "safeguard" designed by FlowyAIPC specifically for weaker or resource-constrained local models. Its core functions include:

    • Remove heavy tools: When enabled, FlowyAIPC automatically removes three tools that require high reasoning capability in each conversation round: browser (browser automation), cron (scheduled tasks), message (message sending)
    • Reduce prompt noise: Decreases system-level tool definitions, lowers the "noise" ratio in the prompt, allowing the model to focus more on core tasks
    • Improve response stability: Prevents weaker models from hallucinating or entering infinite loops due to inability to correctly understand complex tool parameters
    • Save tokens and memory: Reduces context window occupied by tool descriptions, lowering credit consumption and local model pressure

    2. What are the limitations?

    • Missing functionality: Once enabled, the Agent will completely lose the ability to browse the web and exchange messages (but network scraping is still possible), schedule tasks, and actively send messages
    • Non-intelligent trimming: This is a global configuration that does not dynamically adjust based on the actual needs of the current conversation, potentially disabling useful features for simple queries
    • Still requires prompt optimization: Enabling Lite Mode alone is not enough; the System Prompt and group chat rules also need to be streamlined

    3. Scenario Recommendations

    Recommended to enable:

    • Using 7B/8B or smaller parameter models (e.g., Qwen2.5-7B-Instruct, Llama-3-8B)
    • Pure code generation/text processing tasks, only requiring basic capabilities like file read/write and terminal commands
    • Hardware resource bottleneck when GPU VRAM usage is near its limit
    • Sub-agents executing independent atomic tasks that do not need to be aware of the global environment

    Recommended to disable:

    • Tasks that explicitly require accessing online documentation, scraping web data, setting reminders, or pushing messages to external sources

    Steps:

    1. Click the "flask icon" in the top-right corner of the conversation interface to enable Local Model Lite Mode

    Snipaste_2026-06-01_14-29-07

    1. The icon turns blue to indicate successful activation (gateway may restart during the process, please be patient)

    Snipaste_2026-06-01_14-29-18