specification: FinOps Framework
specificationVersion: '1.0'
schema: https://www.finops.org/framework/
provider: Haystack / deepset
providerId: haystack-ai
created: '2026-06-20'
modified: '2026-06-20'
reconciled: false
tags:
- AI
- LLM
- RAG
- Open Source
- FinOps
- Cost Management
- FOCUS
description: >-
  FinOps view of running Haystack and the deepset AI Platform. The open-source
  Haystack framework and self-hosted Hayhooks carry no deepset license fee
  (Apache-2.0); the real spend is the compute you run them on plus the
  per-token/per-call costs of the LLM, embedding, and vector-store providers your
  pipelines call. The hosted deepset AI Platform (deepset Cloud) is billed under
  a sales-led subscription/enterprise agreement, with cost dimensions across
  workspaces, deployed pipelines, indexed files, and search volume. Specific
  rates are not published and are not reconciled here.
notes: >-
  deepset Cloud is quotation-based; obtain a cost breakdown from deepset sales for
  reconciliation. For self-hosted Haystack, model the downstream model/embedding/
  vector-store provider bills and your own infrastructure as the primary cost.
sources:
- https://www.deepset.ai/pricing
- https://github.com/deepset-ai/haystack
- https://docs.cloud.deepset.ai/reference/api-overview
- https://focus.finops.org/focus-specification/v1-3/
alignedWith:
  framework: FinOps Foundation Framework
  frameworkUrl: https://www.finops.org/framework/
  dataSpec: FOCUS
  dataSpecVersion: '1.3'
  dataSpecUrl: https://focus.finops.org/focus-specification/v1-3/
publisherName: deepset
serviceCategory: AI and Machine Learning
billingModel:
  pricingCategory: Subscription-Based
  billingFrequency: Annual
  billingCurrency: USD
  chargeCategories:
  - Usage
  - Purchase
focusColumns:
  ServiceName: deepset AI Platform
  ServiceCategory: AI and Machine Learning
  ProviderName: deepset
  PublisherName: deepset
  InvoiceIssuerName: deepset
  BillingCurrency: USD
  ChargeCategory: Purchase
  PricingCategory: Subscription-Based
meters:
- name: deployed_pipelines
  description: Pipelines deployed and kept running in deepset Cloud workspaces.
  unit: pipelines
  aggregation: max
  dimensions:
  - workspace
  - pipeline
- name: search_requests
  description: Queries run against deployed pipelines via the hosted search endpoint.
  unit: requests
  aggregation: sum
  dimensions:
  - workspace
  - pipeline
- name: indexed_files
  description: Files uploaded and indexed into workspaces for retrieval.
  unit: files
  aggregation: sum
  dimensions:
  - workspace
- name: downstream_model_tokens
  description: >-
    Tokens billed by the LLM/embedding providers that Haystack pipelines call
    (self-hosted or hosted). Not a deepset charge but the dominant variable cost.
  unit: tokens
  aggregation: sum
  dimensions:
  - provider
  - model
- name: compute_hours
  description: Self-hosted compute running Haystack and Hayhooks deployments.
  unit: hours
  aggregation: sum
  dimensions:
  - environment
principles:
- name: Visibility
  description: >-
    Separate the free framework from real spend - track downstream model/embedding
    provider bills and infrastructure for self-hosted Haystack, and the deepset
    Cloud subscription for the hosted platform.
- name: Allocation
  description: Tag pipelines and workspaces per team/workload to map deepset Cloud and downstream model spend to cost centers.
- name: Optimization
  description: >-
    Right-size retrievers and generators, cache embeddings, choose smaller/cheaper
    models where quality allows, and undeploy idle deepset Cloud pipelines.
- name: Accountability
  description: Assign owners per workspace/pipeline and review subscription plus downstream model spend against budget.
maintainers:
- FN: Kin Lane
  email: kin@apievangelist.com