Elastic-Claude

Local search infrastructure for project knowledge.

CLI Commands

Add an entry

# From file path (preferred - Rust reads the file)
elastic-claude add -t <type> -p <file_path> [-m '<json_metadata>']

# Inline content (for small content)
elastic-claude add -t <type> -c "<content>" [-m '<json_metadata>']

# From stdin
cat <file> | elastic-claude add -t <type> [-m '<json_metadata>']

Arguments:

• -t, --entry-type

  : Entry type (e.g., "document", "chat", "code")

• -p, --path

  : Read content from file (also sets file_path in DB)

• -c, --content

  : Inline content (conflicts with -p)

• -m, --metadata

  : JSON metadata (optional)

Example:

elastic-claude add -t document -p /path/to/file.md -m '{"project": "my-project", "title": "My Doc", "category": "docs"}'

Search entries

elastic-claude search "<query>"

Returns matching entries with snippets, scores, and metadata.

Save current chat session

# Get path to current chat file
elastic-claude current-chat --path-only

# Ingest current chat with metadata (always include project)
elastic-claude current-chat -m '{"project": "my-project", "title": "Session title", "tags": ["topic1", "topic2"]}'

Auto-detects the current Claude session and ingests the chat file.

Get entry by ID

# Get full entry details
elastic-claude get <id>

# Get just the content
elastic-claude get <id> --content-only

# Show tsvector tokens (for debugging search)
elastic-claude get <id> --tsv

Workflow for Ingesting Documents

When asked to ingest files:

1. Determine the project name from the current working directory or ask the user
2. Read each file's content
3. Extract metadata from content and path:
   • project: always include - the project/repo name
   • title: from first heading or filename
   • category: from directory path
   • tags: from content keywords
4. Call

   elastic-claude add

   for each file

Example for a markdown file:

elastic-claude add -t document -p /path/to/file.md -m '{"project": "CQR", "title": "Extracted Title", "category": "documentation"}'

Workflow for Searching

When asked to find information:

1. Run

   elastic-claude search "<relevant keywords>"

2. Review the results (snippets, file paths, metadata)
3. If needed, read the full files for more context
4. Summarize findings for the user

Schema Reference

See references/schema.md for database schema details.

Entry types:

• document

  : Markdown docs, READMEs, etc.

• chat

  : Chat session transcripts

• code

  : Code snippets or files