436 Assistant Context 5ce22085

Before proceeding, ALWAYS INVOKE the pinecone:assistant skill before commencing workflow. This will allow correct resolution of plugin root directory for scripts being run.

Script paths are relative to the plugin root directory. Run scripts with: uv run skills/assistant/scripts/script_name.py [arguments]

Overview

Retrieve relevant context snippets from a Pinecone Assistant's knowledge base without generating a full chat response. This is useful for:

• Quick knowledge base searches
• Finding specific information in uploaded documents
• Debugging what the assistant knows
• Building custom RAG workflows

Workflow

1. Parse Arguments:

   • assistant: Name of the assistant (required)
   • query: Search query text (required)
   • top_k: Number of snippets to return (optional, default: 5, max: 16)
   • snippet_size: Maximum tokens per snippet (optional, default: 2048)
   • json: Output in JSON format (optional flag)

2. If Arguments Missing:

   • If no assistant: List assistants and use AskUserQuestion to let user select
   • If no query: Prompt user for their search text

3. Execute Context Script:

   uv run skills/assistant/scripts/context.py \
     --assistant "assistant-name" \
     --query "search text" \
     --top-k 5 \
     --snippet-size 2048
   

4. Display Results:

   • Context snippets displayed in panels with:
     • File name and page numbers
     • Relevance scores (0-1)
     • Actual snippet content (truncated if > 500 chars)
     • Relevance percentage

5. Offer Next Actions:

   • Suggest using /pinecone:assistant-chat to ask a full question
   • Offer to refine the search with a different query
   • Suggest uploading more documents if no results found

Example Usage

Basic context retrieval:

/pinecone:assistant-context assistant docs-qa query "authentication methods"

With custom top-k:

/pinecone:assistant-context assistant support-bot query "reset password" --top-k 10

JSON output:

/pinecone:assistant-context assistant my-docs query "API endpoints" --json

Interactive mode:

• User invokes: /pinecone:assistant-context
• Claude lists assistants and asks user to select
• Claude prompts for the search query
• Claude executes the script and displays results

Use Cases

Quick Lookup: Find specific information without needing a full conversational response

Query: "pricing tiers"
Result: Returns relevant sections from pricing documentation

Debugging Knowledge: Check what documents the assistant has about a topic

Query: "deployment"
Result: Shows all relevant deployment-related content

Pre-filtering: See what context exists before asking a question

Query: "authentication"
Result: Preview relevant docs, then ask detailed question in chat

Custom Workflows: Get raw context for building your own RAG pipelines

Script Location

skills/assistant/scripts/context.py

Troubleshooting

Error: PINECONE_API_KEY not set

• Remind user to export PINECONE_API_KEY before starting Claude Code

Error: Assistant not found

• List available assistants
• Check for typos in assistant name

No results found

• Assistant may not have relevant documents uploaded
• Try broader search terms
• Suggest uploading relevant files with /pinecone:assistant-upload
• Offer to show what documents are in the assistant

Error: context method not available

• Check that you're using Pinecone SDK v8.0.0+ which includes assistant.context() support
• The context API requires API version 2025-04 or later for full parameter support
• If the method doesn't exist, your SDK may need updating: pip install --upgrade pinecone
• Fallback suggestion: Use /pinecone:assistant-chat instead to get responses with citations

Context vs Chat

Help users understand the difference:

Use Context when:

• You want quick snippets without generated responses
• You're debugging the knowledge base
• You need raw source material
• You're building custom workflows

Use Chat when:

• You want conversational Q&A
• You need synthesized answers from multiple sources
• You want citations with the response
• You're having a back-and-forth conversation

Interpreting Results

Guide users on reading context results:

• Score: Higher scores (closer to 1.0) mean more relevant
• File Name: Which document the snippet came from
• Page Number: Exact page reference for PDFs
• Text: The actual content snippet

Low scores (<0.5) may indicate:

• Weak match to query
• Assistant needs more relevant documents
• Query too broad or too specific

Related Commands

• /pinecone:assistant-chat - Ask questions and get full responses
• /pinecone:assistant-upload - Add more documents to improve results
• List assistants: Run uv run skills/assistant/scripts/list.py