Skip to main content

​
The System Engine Room

The Advanced tab houses the primary architectural knobs for the MemoryBox application ecosystem. The configurations are partitioned across distinct functional layers, enabling you to bring your own specialized LLM endpoints, audit routing networks, define model visibility states, and modify core text vectorization parameters.

​
🚀 Page 1: Endpoint Integration & Network Gateways

The initial view of the Advanced pane handles data connection parameters, system routing loops, and credential verification checks.
Advanced custom endpoint and routing configuration settings

The primary Advanced panel interface tracking local model additions and system gateway paths.

​
1. Experimental Features Toggle

Located at the very top of the Advanced pane, the Experimental features toggle activates near-stable pilot capabilities (such as the App Builder). Toggling this setting takes effect immediately and persists across application restarts.

​
2. Bring Your Own Model

The Bring your own model container enables the onboarding of custom execution targets. This module accepts any standard OpenAI-compatible server layout, passing keys securely to your operating system’s native credential store.
  • Display name: A personalized identifier text string used to name the model within your chat workspace.
  • Base URL: The endpoint target network address (e.g., https://...).
  • Upstream model id: The exact model string identifier required by the provider’s execution API.
  • API key: The authorization token string passed safely to local on-device hardware storage.
  • Add model Action Button: Registers the customized engine, immediately loading it into your available model repository list.

​
3. Platform Gateway Routing

The Platform Gateway container displays active localization routing nodes and status (e.g., Current Route: US):
  • Automatic gateway routing check: Periodically audits connection pathways while the app runs. This feature ensures smooth transitions during network events, such as turning on a VPN or changing corporate proxy settings.
  • Check interval Dropdown: Controls the baseline audit frequency, defaulting to Every 5 minutes (default).
  • Check now Button: Triggers an immediate manual network path discovery cycle to re-route platform requests instantly.

​
👥 Platform Models Deployment Matrix

Clicking the Platform models accordion layout element expands the interface to reveal your upstream cloud intelligence matrix.
Expanded platform model visibility toggle grid

The global Platform Models visibility dashboard tracking live toggle states across providers.

Instead of cluttering your main workspace interfaces with hundreds of irrelevant endpoints, this layout groups models by provider. Toggling the Show in picker switch determines whether that individual expert appears inside your chat window composer dropdown menus. Four core flagship models start enabled out of the box, while remaining variants start hidden.

​
Supported Model Roster

The platform explicitly supports the following baseline model definitions:

​
⚙️ Page 2: Memory Backend Runtime Configuration

Scrolling down to the second segment exposes the core variables governing how data chunks are prepared, indexed, or segmented.
Memory backend vector and segmentation config view

The advanced indexing parameter cards used to alter the system runtime environment.

​
1. General & Search Heuristics

  • Embedding Chunk Max Bytes: Dictates the hard memory byte ceiling applied when segmenting target document inputs into discrete vector tokens. Increasing this parameter captures larger continuous file sections, while lower integers increase total text processing efficiency. Defaults to 8000 bytes.
  • Log Level: Modifies tracking verbosity configurations for system troubleshooting logs.
  • Search Method: Sets the target algorithmic model indexing style used to traverse local vectors.
  • Rerank Mode: Determines the logic applied to prioritize highly relevant semantic matches. Setting this parameter to local processes computations natively on-device to enforce user privacy data boundaries.
  • Rerank Model Directory: Specifies the path location link tracking where local ranking files are stored. Click Browse Model Directory to select a folder location.
  • Rerank Timeout (ms): Sets the absolute millisecond cutoff duration (defaults to 1500) the index engine waits before abandoning a search rank operation.
  • Minimum Vector Score: Defines the baseline threshold score (defaults to 0.7) required for semantic vector match retrieval.

​
2. Session Episode Segmentations

These parameters adjust the criteria MemoryBox uses to analyze continuous local logs and parse them into historical chat topics:
  • Session Episode Segment Max Gap (s): Sets the maximum allowable second gap window (defaults to 300) between incoming entries before the core system separates the conversation into a new topic block.
  • Session Episode Segment Max Messages: Sets the total maximum message threshold counts (defaults to 12) allowed within an individual indexed segment block.

​
💾 Saving Configuration Adjustments

Because modifications to this panel alter core system behaviors, updates are not applied dynamically upon clicking field entries:
  • Load Current Config Button: Refreshes the active screen view to read parameters currently running in the local app environment.
  • Reset To Loaded Button: Aborts pending modifications, reverting form fields back to your last saved state.
  • Apply Config Button: Writes all changes to the server database backend and updates the model picker roster.
Applying Changes: Navigating away from this tab, changing navigation links, or closing the main settings modal prior to clicking Apply Config will immediately discard all adjustments.