System Settings and Model Discovery

The Settings view allows administrators to manage vector database engines, embedding models, LiteLLM proxy discovery, file thresholds, and domain-level Git credentials.

---

1. Vector Database Management

ContextCortex supports dynamic switching between vector database engines:

• Supported Backends:
  • Qdrant (Recommended): High-performance embedded or remote vector database with native sparse BM25 payload support.
  • pgvector: PostgreSQL 16 relational database with pgvector extensions.
  • ChromaDB: Lightweight local or distributed vector store.
• Connection Testing: Click Test Connection to verify host connectivity, port reachability, and collection state before applying changes.
• Operating Modes: Toggle between Embedded (zero external dependencies) and Remote (external cluster or container).

---

2. LiteLLM Dynamic Model Discovery

When connected to an upstream LiteLLM proxy (LITELLM_URL), ContextCortex queries the model capabilities matrix dynamically and separates models into specialized categories:

• Embedding Models: Filtered by embedding capabilities (e.g. BAAI/bge-small-en-v1.5, text-embedding-3-small).
• Vision OCR Models: Multi-modal vision models used to perform automated OCR on diagrams and scanned pages in PDF uploads (e.g. gemini-2.5-flash, qwen3-vl).
• Chat & Summarization Models: Reasoning models used to generate structured executive summaries for large files (e.g. gemini-2.5-flash, gpt-4o-mini).

---

3. Files & Large File Summarization Settings

Configure file access ceilings and LLM summarization thresholds:

Setting	Default	Description
Enable Large File Summarization	true	When enabled, files larger than the threshold are auto-summarized rather than omitted or raw-chunked.
Auto-Summarize Threshold (KB)	500 KB	File size threshold beyond which ingestion triggers automated LLM summarization.
Maximum File Size (MB)	10 MB	Absolute hard ceiling above which files are skipped entirely.
Max Lines Per Read	2,000 lines	Line range ceiling returned by the read_file tool to protect agent context windows.
Summary Chat Model	gemini-2.5-flash	Dedicated LLM model identifier used for generating executive file summaries.

---

4. Git Host Credential Vault

Store domain-level credentials for internal and enterprise Git instances:

1. Scroll to the Git Host Credentials table.
2. Click + Add Host Credential.
3. Host Domain: Enter the domain name (e.g. gitlab.internal.company.com or git.selfhosted.net).
4. Auth User & Token: Enter the service user and personal access token (PAT).
5. All repositories matching that host domain inherit these credentials automatically without needing individual tokens.

---

Next Steps

• Inspect system health and operational traces in Diagnostics and System Logs.
• Customize dashboard palettes in Appearance and Theme Customization.