Skip to main content

Basic Settings and Custom AI Configuration

After installing and launching Memfit, configure two areas first:

  1. Basic settings: control tool permissions, manual review, local file access, retries, and task limits.
  2. Custom AI configuration: add your own AI provider, model, API Key, BaseURL / Endpoint, and routing policy.

These settings determine whether Memfit can complete tasks safely and reliably. For first-time use, follow the recommended setup below, then adjust the details for your workflow.

【Image: Memfit settings entry, showing the configuration panel and AI model configuration entry】

Open Settings

In the Memfit main interface:

  1. Click Settings or the configuration button in the upper-right corner.
  2. Adjust basic settings in the chat configuration panel.
  3. Open AI Model Configuration to add or select models.

If you want to start quickly, keep Use default system AI configuration enabled and review the permission controls. If you need to use your own API Key or private model service, continue with custom AI configuration.

Basic Settings

Basic settings control what the AI Agent can do, when it needs user approval, and how it retries failed tasks.

Permission Controls

SettingRecommendationPurpose
Enable system file operation permissionEnable as neededAllows AI to read task files and generate or modify files. Enable it for real project work; disable it for chat-only demos.
Disable ToolsKeep disabledWhen enabled, AI cannot call external tools and Memfit behaves like a pure chat assistant.
Disable tool runtime AI reviewKeep disabledSkips AI review during tool execution. This can improve speed but reduces automatic risk checks.
Review ruleChoose Manual firstControls high-risk action review: Manual asks the user, AI lets AI decide, and Yolo favors automatic execution.
Risk thresholdDefault 0.5Actions below the threshold can be approved automatically; higher-risk actions are escalated to manual review.

【Image: Basic settings panel, highlighting file permission, Review rule, risk threshold, and Disable Tools】

Human Interaction Controls

SettingRecommendationPurpose
Disable human interactionEnabled by defaultPrevents AI from repeatedly asking questions during execution.
Allow human interaction during planningEnabled by defaultAllows AI to ask clarifying questions while preparing a task plan.
Planning interaction countDefault 3Limits how many questions AI can ask during planning.
User interaction limitConfigure as neededLimits total user prompts during a task.

Task and Context Limits

SettingDefaultPurpose
AI call retry count3Retries unstable remote AI calls.
AI transaction retry count5Retries poor-quality or failed task transactions.
ReAct iteration limit100Prevents tasks from running indefinitely.
Timeline item limit100Limits retained timeline entries.
Timeline content size60 KBLimits timeline context size.
Token pressure threshold40 KWarns or compresses context when token pressure is high.

Keep defaults for everyday use. Increase the ReAct iteration limit only when complex tasks stop too early.

Custom AI Configuration

Memfit organizes AI models by role:

Model TypePurposeSuitable Models
High-quality modelPlanning, reasoning, code analysis, security analysisStronger reasoning models
Lightweight modelFrequent low-cost classification, summarization, routingFast, low-cost models
Vision modelScreenshots, UI, charts, imagesModels with image input support

Add an AI Model

In AI Model Configuration, click Add and fill in:

FieldHow to Fill
ProviderSelect a built-in provider or enter a custom OpenAI-compatible provider name.
Model typeChoose High-quality model, Lightweight model, or Vision model.
APItypeSelect chat_completions or responses.
Model nameChoose or enter a model name, such as gpt-4o, gpt-4o-mini, or deepseek-chat.
ApiKeyEnter the provider API Key / Token. Some built-in free services may not require a real key.

【Image: Add AI model modal, highlighting Provider, Model type, APItype, Model name, and ApiKey】

Configure BaseURL or Endpoint

For private deployments, proxy services, or OpenAI-compatible gateways, expand Advanced Configuration:

Advanced SettingUse Case
BaseURLUse for standard OpenAI-style paths, such as https://api.openai.com/ or https://api.example.com/v1.
Enable EndpointEnable when the service does not use standard OpenAI paths.
EndpointEnter the full request URL, such as https://api.example.com/v1/chat/completions.
Proxy addressEnter a proxy such as http://127.0.0.1:7890 when required.
HeaderAdd extra request headers required by an enterprise gateway.

【Image: Advanced AI model settings, highlighting BaseURL, Endpoint, proxy, and Header】

Select Active Models

After adding models:

  1. Expand High-quality model, Lightweight model, or Vision model.
  2. Find the model you added.
  3. Select it as the active model.
  4. Configure at least one High-quality model. Add a Lightweight model to reduce high-frequency task cost.

Choose Calling Mode

ModeDescriptionBest For
AutoAutomatically selects a model by task contentRecommended default
PerformancePrioritizes high-quality modelsQuality-first reasoning
CostPrioritizes lightweight modelsHigh-frequency, budget-sensitive tasks
BalanceBalances quality, speed, and costEveryday use

Fallback Strategy

Disable fallback controls whether Memfit can switch models when the current model is unavailable:

  • Keep disabled: Memfit can try another tier, such as a lightweight model, when the high-quality model fails.
  • Enable it: model failure stops the task, suitable for strict audit or fixed-model workflows.
  1. Enable system file operation permission.
  2. Set Review rule to Manual.
  3. Keep Disable Tools off.
  4. Keep Use default system AI configuration on, or add your own high-quality model.
  5. Set calling mode to Auto.
  6. Keep Disable fallback off.

Then return to the main interface and try:

Summarize the current project structure and identify the main entry files.

If Memfit can create a plan, request permissions when needed, and produce a result, the basic settings and AI configuration are ready.

Next Steps

  1. Hello, Memfit! - Start your first AI Agent experience
  2. Advanced Usage - Inspect usage, control cost, and customize capabilities
Published: