specificationVersion: "1.0"
id: cognee-finops
name: Cognee FinOps Framework
description: >
  FinOps guidance for managing and optimizing spend on the Cognee AI memory and knowledge
  graph platform. Cognee costs span three dimensions: (1) the Cognee subscription plan
  for managed cloud hosting, (2) pass-through LLM inference costs from the underlying
  language model used during the cognify pipeline, and (3) infrastructure costs for vector
  and graph storage. The cost calculator at https://www.cognee.ai/cost-calculator helps
  estimate all three components before committing to a plan.
url: https://www.cognee.ai/pricing

focus:
  - dimension: ProviderCost
    description: Cognee subscription plan charges (Free / Developer / Cloud Team / Enterprise)
    metrics:
      - name: MonthlySubscription
        unit: USD/month
        values:
          Free: 0
          Developer: 35
          CloudTeam: 200
          Enterprise: custom

  - dimension: UsageCost
    description: >
      Top-up document pack purchases when the plan's base document allowance is exhausted.
      Charged per top-up purchased, not metered automatically.
    metrics:
      - name: DocumentTopUp1k
        unit: USD/pack
        value: 35
        coverage: 1000 docs / 1 GB
      - name: DocumentTopUp3k
        unit: USD/pack
        value: 100
        coverage: 3000 docs / 3 GB
      - name: DocumentTopUp15k
        unit: USD/pack
        value: 750
        coverage: 15000 docs / 15 GB

  - dimension: PassThroughCost
    description: >
      LLM inference costs billed directly by the chosen language model provider during
      the cognify extraction and summarization pipeline. These are not collected by
      Cognee; they are paid separately to the LLM provider (e.g., OpenAI, Anthropic,
      DeepSeek). Cognee supports 29 LLM providers, with per-million-token rates ranging
      from ~$0.28/$0.42 (DeepSeek V3) to $15/$75 (Claude Opus).
    metrics:
      - name: EmbeddingCost
        provider: OpenAI
        model: text-embedding-3-small
        unit: USD/million tokens
        inputRate: 0.02
      - name: EmbeddingCostLarge
        provider: OpenAI
        model: text-embedding-3-large
        unit: USD/million tokens
        inputRate: 0.13
      - name: GraphExtractionCost
        description: Typical LLM call cost for graph entity/relationship extraction
        unit: USD/million tokens
        range:
          low: 0.28
          high: 15.00

  - dimension: InfrastructureCost
    description: >
      Monthly storage costs for the vector database and graph database backends.
      Cognee supports multiple backends; costs vary by provider and tier chosen.
    metrics:
      - name: VectorStoreFree
        provider: Qdrant Free
        unit: USD/month
        value: 0
      - name: VectorStoreStandard
        provider: Qdrant Standard
        unit: USD/month
        value: 10
      - name: GraphDBFree
        provider: Neo4j AuraDB Free
        unit: USD/month
        value: 0
      - name: GraphDBEnterprise
        provider: Neo4j AuraDB Enterprise
        unit: USD/month
        range:
          low: 65
          high: 146

optimization:
  - strategy: ChooseEfficientLLM
    description: >
      Select a cost-efficient LLM for the cognify pipeline (e.g., DeepSeek V3 or
      GPT-4o Mini) for bulk ingestion tasks. Reserve premium models (Claude Opus,
      GPT-4o) for high-stakes retrieval or reasoning workloads.
  - strategy: BatchIngestion
    description: >
      Batch document ingestion into larger jobs to reduce per-call overhead and
      maximize throughput under the included API call quota before purchasing top-up packs.
  - strategy: SelfHostForHighVolume
    description: >
      At high document volumes, self-hosting Cognee on Modal, Railway, or Fly.io may
      reduce per-document costs by eliminating subscription markup and giving direct
      control over LLM and storage provider selection.
  - strategy: UseCostCalculator
    description: >
      Use Cognee's cost calculator (https://www.cognee.ai/cost-calculator) to estimate
      total cost across LLM, embedding, compute, and storage before scaling ingestion.
  - strategy: MonitorAPICallUsage
    description: >
      Track monthly API call consumption against the 10,000-call plan limit to avoid
      unexpected top-up pack purchases. Monitor via the Cognee platform dashboard.