Basic Settings and Custom AI Configuration
After installing and launching Memfit, configure two areas first:
- Basic settings: control tool permissions, manual review, local file access, retries, and task limits.
- 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:
- Click Settings or the configuration button in the upper-right corner.
- Adjust basic settings in the chat configuration panel.
- 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
| Setting | Recommendation | Purpose |
|---|---|---|
| Enable system file operation permission | Enable as needed | Allows AI to read task files and generate or modify files. Enable it for real project work; disable it for chat-only demos. |
| Disable Tools | Keep disabled | When enabled, AI cannot call external tools and Memfit behaves like a pure chat assistant. |
| Disable tool runtime AI review | Keep disabled | Skips AI review during tool execution. This can improve speed but reduces automatic risk checks. |
| Review rule | Choose Manual first | Controls high-risk action review: Manual asks the user, AI lets AI decide, and Yolo favors automatic execution. |
| Risk threshold | Default 0.5 | Actions 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
| Setting | Recommendation | Purpose |
|---|---|---|
| Disable human interaction | Enabled by default | Prevents AI from repeatedly asking questions during execution. |
| Allow human interaction during planning | Enabled by default | Allows AI to ask clarifying questions while preparing a task plan. |
| Planning interaction count | Default 3 | Limits how many questions AI can ask during planning. |
| User interaction limit | Configure as needed | Limits total user prompts during a task. |
Task and Context Limits
| Setting | Default | Purpose |
|---|---|---|
| AI call retry count | 3 | Retries unstable remote AI calls. |
| AI transaction retry count | 5 | Retries poor-quality or failed task transactions. |
| ReAct iteration limit | 100 | Prevents tasks from running indefinitely. |
| Timeline item limit | 100 | Limits retained timeline entries. |
| Timeline content size | 60 KB | Limits timeline context size. |
| Token pressure threshold | 40 K | Warns 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 Type | Purpose | Suitable Models |
|---|---|---|
| High-quality model | Planning, reasoning, code analysis, security analysis | Stronger reasoning models |
| Lightweight model | Frequent low-cost classification, summarization, routing | Fast, low-cost models |
| Vision model | Screenshots, UI, charts, images | Models with image input support |
Add an AI Model
In AI Model Configuration, click Add and fill in:
| Field | How to Fill |
|---|---|
| Provider | Select a built-in provider or enter a custom OpenAI-compatible provider name. |
| Model type | Choose High-quality model, Lightweight model, or Vision model. |
| APItype | Select chat_completions or responses. |
| Model name | Choose or enter a model name, such as gpt-4o, gpt-4o-mini, or deepseek-chat. |
| ApiKey | Enter 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 Setting | Use Case |
|---|---|
| BaseURL | Use for standard OpenAI-style paths, such as https://api.openai.com/ or https://api.example.com/v1. |
| Enable Endpoint | Enable when the service does not use standard OpenAI paths. |
| Endpoint | Enter the full request URL, such as https://api.example.com/v1/chat/completions. |
| Proxy address | Enter a proxy such as http://127.0.0.1:7890 when required. |
| Header | Add 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:
- Expand High-quality model, Lightweight model, or Vision model.
- Find the model you added.
- Select it as the active model.
- Configure at least one High-quality model. Add a Lightweight model to reduce high-frequency task cost.
Choose Calling Mode
| Mode | Description | Best For |
|---|---|---|
| Auto | Automatically selects a model by task content | Recommended default |
| Performance | Prioritizes high-quality models | Quality-first reasoning |
| Cost | Prioritizes lightweight models | High-frequency, budget-sensitive tasks |
| Balance | Balances quality, speed, and cost | Everyday 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.
Recommended First Setup
- Enable system file operation permission.
- Set Review rule to
Manual. - Keep Disable Tools off.
- Keep Use default system AI configuration on, or add your own high-quality model.
- Set calling mode to Auto.
- 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
- Hello, Memfit! - Start your first AI Agent experience
- Advanced Usage - Inspect usage, control cost, and customize capabilities