DrSai Knowledge Base Query

DrSai Desktop manages knowledge bases through a local Gateway process (127.0.0.1:28643). The script talks to the Gateway REST API and supports both KB types via the same search-preview endpoint:

Type	Description
local-files	Local workspace files indexed in DrSai Desktop
ragflow	Remote RAGFlow datasets

When to Use

Activate this skill when the user asks to:

• Search / retrieve / read from "知识库" (knowledge base)
• Look up information in RAGFlow datasets or local indexed documents
• Find content in saved knowledge sources

How to Use

Run the bundled script scripts/query_kb.py with the Codex primary Python runtime.

List available knowledge bases:

python scripts/query_kb.py --list

Search all knowledge bases:

python scripts/query_kb.py --query "your search term"

Filter by type:

python scripts/query_kb.py --query "keyword" --kb-type local-files
python scripts/query_kb.py --query "keyword" --kb-type ragflow

Filter by KB name:

python scripts/query_kb.py --query "keyword" --kb-name <display_name>

Search an unregistered local folder:

If the local folder is not yet registered in DrSai Desktop:

python scripts/query_kb.py --query "keyword" --local-path /path/to/docs

Or set the DRSAI_LOCAL_KB environment variable.

Output

Each result includes:

• KB name and type (local-files / ragflow)
• Source document path within the KB
• Relevance score (0-1)
• Content snippet (up to 600 chars of the matched chunk)

Important Notes

• DrSai Desktop must be running (Gateway at 127.0.0.1:28643)
• The instance token is auto-detected from ~/.drsai-prod/runtime/instance-token
• Both KB types use the same search-preview API
• The script auto-discovers the DrSai home directory (.drsai-prod / .drsai / .drsai-dev)
• If no knowledge bases are configured, guide the user to add one in DrSai Desktop

Architecture

query_kb.py → Gateway REST (127.0.0.1:28643)
  ├── GET    /v1/config/knowledge-bases                  (list KBs)
  └── POST   /v1/config/knowledge-bases/{id}/search-preview  (search)
        → returns evidence[] with content, score, source