Control access to knowledge base collections with the AI RAG Injector plugin

echo '
_format_version: "3.0"
plugins:
  - name: ai-proxy-advanced
    config:
      targets:
      - route_type: llm/v1/chat
        auth:
          header_name: Authorization
          header_value: Bearer ${{ env "DECK_OPENAI_API_KEY" }}
        model:
          provider: openai
          name: gpt-4o
          options:
            max_tokens: 512
            temperature: 1.0
' | deck gateway apply -

echo '
_format_version: "3.0"
plugins:
  - name: key-auth
    config:
      key_names:
      - apikey
      key_in_header: true
      key_in_query: true
      hide_credentials: true
' | deck gateway apply -

echo '
_format_version: "3.0"
consumer_groups:
  - name: public
  - name: finance
  - name: executive
  - name: contractor
' | deck gateway apply -

echo '
_format_version: "3.0"
consumers:
  - username: cfo
    custom_id: cfo-001
    groups:
    - name: finance
    - name: executive
    keyauth_credentials:
    - key: cfo-key
  - username: financial-analyst
    custom_id: analyst-001
    groups:
    - name: finance
    keyauth_credentials:
    - key: analyst-key
  - username: contractor-dev
    custom_id: contractor-001
    groups:
    - name: contractor
    keyauth_credentials:
    - key: contractor-key
  - username: public-user
    custom_id: public-001
    groups:
    - name: public
    keyauth_credentials:
    - key: public-key
' | deck gateway apply -

echo '
_format_version: "3.0"
plugins:
  - name: ai-rag-injector
    id: b924e3e8-7893-4706-aacb-e75793a1d2e9
    config:
      embeddings:
        auth:
          header_name: Authorization
          header_value: Bearer ${{ env "DECK_OPENAI_API_KEY" }}
        model:
          provider: openai
          name: text-embedding-3-large
      vectordb:
        strategy: redis
        dimensions: 3072
        distance_metric: cosine
        redis:
          host: "${{ env "DECK_REDIS_HOST" }}"
          port: 6379
      inject_template: |
        Use the following context to answer the question. If the context doesnt contain relevant information, say so.
        Context:
        <CONTEXT>
        Question: <PROMPT>
      inject_as_role: system
      consumer_identifier: consumer_group
      global_acl_config:
        allow:
        - public
        deny: []
      collection_acl_config:
        public-docs:
          allow: []
          deny: []
        finance-reports:
          allow:
          - finance
          - executive
          deny:
          - contractor
        executive-confidential:
          allow:
          - executive
' | deck gateway apply -

cat > ingest-collection.py << 'EOF'
#!/usr/bin/env python3
import requests
import json

BASE_URL = "http://localhost:8001/ai-rag-injector/b924e3e8-7893-4706-aacb-e75793a1d2e9/ingest_chunk"

chunks = [
    {
        "content": "Public Investor FAQ: Our fiscal year ends December 31st. Quarterly earnings calls occur in January, April, July, and October. All public filings are available on our investor relations website. For questions, contact investor.relations@company.com.",
        "metadata": {
            "collection": "public-docs",
            "source": "website",
            "date": "2024-01-15T00:00:00Z",
            "tags": ["public", "investor-relations", "faq"]
        }
    },
    {
        "content": "Q4 2024 Financial Results: Revenue increased 15% year-over-year to $2.3B. Operating margin improved to 24%, up from 21% in Q3. Key drivers included strong enterprise sales and improved operational efficiency.",
        "metadata": {
            "collection": "finance-reports",
            "source": "internal",
            "date": "2024-10-14T00:00:00Z",
            "tags": ["finance", "quarterly", "q4", "2024"]
        }
    },
    {
        "content": "Q3 2024 Financial Results: Revenue reached $2.0B with 12% year-over-year growth. Operating margin held steady at 21%. International markets contributed 35% of total revenue.",
        "metadata": {
            "collection": "finance-reports",
            "source": "internal",
            "date": "2024-07-15T00:00:00Z",
            "tags": ["finance", "quarterly", "q3", "2024"]
        }
    },
    {
        "content": "2023 Annual Report: Full-year revenue totaled $7.8B, representing 18% growth. The company expanded into three new markets and launched five major product updates. Board approved $500M share buyback program.",
        "metadata": {
            "collection": "finance-reports",
            "source": "internal",
            "date": "2023-12-31T00:00:00Z",
            "tags": ["finance", "annual", "2023"]
        }
    },
    {
        "content": "Historical Data Archive: Q2 2022 revenue was $1.5B with 8% growth. This data is retained for historical analysis but may not reflect current business conditions or reporting standards.",
        "metadata": {
            "collection": "finance-reports",
            "source": "archive",
            "date": "2022-06-15T00:00:00Z",
            "tags": ["finance", "quarterly", "q2", "2022", "archive"]
        }
    },
    {
        "content": "CONFIDENTIAL - M&A Discussion: Preliminary valuation for Target Corp acquisition ranges from $400M-$500M. Due diligence reveals strong synergies in enterprise segment. Board vote scheduled for Q1 2025. Legal counsel: Morrison & Associates. Internal deal code: MA-2024-087.",
        "metadata": {
            "collection": "executive-confidential",
            "source": "internal",
            "date": "2024-11-20T00:00:00Z",
            "tags": ["confidential", "m&a", "executive"]
        }
    }
]

def ingest_chunks():
    headers = {
        "Content-Type": "application/json",
        "apikey": "admin-key"
    }

    for i, chunk in enumerate(chunks, 1):
        try:
            response = requests.post(BASE_URL, json=chunk, headers=headers)
            response.raise_for_status()
            print(f"[{i}/{len(chunks)}] Ingested: {chunk['content'][:50]}...")
            print(response.json())
        except requests.exceptions.RequestException as e:
            print(f"[{i}/{len(chunks)}] Failed: {e}")
            if hasattr(e.response, 'text'):
                print(f"  Response: {e.response.text}")

if __name__ == "__main__":
    ingest_chunks()
EOF

python3 ingest-collection.py