name: Unify FinOps Framework
description: >
  FinOps Framework 1.0 FOCUS-aligned cost management data for the Unify LLM
  routing and model gateway platform. Unify's core value proposition is cost
  optimization: the routing layer selects the cheapest provider meeting quality
  and latency thresholds on every prompt, providing an inherent FinOps function.
  Costs are consumption-based, denominated in USD credits, and passed through
  from underlying providers with Unify's routing overhead included.
specificationVersion: '1.0'
framework: FinOps Framework
alignment: FOCUS 1.0
provider: Unify
providerURL: https://unify.ai
costModel:
  type: consumption
  currency: USD
  billingUnit: credits
  description: >
    Unify charges on a credit consumption basis. Credits map to underlying LLM
    API costs (input tokens + output tokens) at the target model and provider,
    plus Unify's routing and infrastructure margin. Users prepay credits via the
    console and are debited per request.
costCategories:
  - name: LLM Token Consumption
    description: >
      Primary cost driver. Input and output tokens processed by the selected LLM
      provider. Cost per token varies by model and provider (e.g., GPT-4o at
      OpenAI is more expensive than Llama-3 at Together AI). Unify's optimizer
      selects cheaper providers that meet defined quality/latency thresholds,
      directly reducing this cost category.
    unit: per token (input + output)
    variability: high
    optimizationStrategies:
      - Use quality-cost-speed optimizer routing to select cheapest qualifying provider
      - Enable response caching to avoid duplicate token charges on repeated prompts
      - Set explicit cost thresholds in routing optimizer parameters
      - Use smaller/cheaper models for classification and routing subtasks
  - name: API Request Overhead
    description: >
      Per-request routing and infrastructure overhead charged by Unify on top of
      underlying provider costs. Covers the routing intelligence, benchmarking
      data freshness, and gateway reliability layer.
    unit: per request
    variability: low
    optimizationStrategies:
      - Batch requests where possible to amortize per-request overhead
      - Use caching layer to serve repeated prompts without re-routing
  - name: Benchmark and Evaluation Costs
    description: >
      Running model comparison benchmarks and custom evaluations on Unify consumes
      credits proportional to the number of model-prompt combinations evaluated.
      Benchmarking multiple providers simultaneously multiplies token costs.
    unit: per benchmark run (per model x prompt)
    variability: medium
    optimizationStrategies:
      - Limit benchmark scope to candidate providers rather than all 100+
      - Cache benchmark results and refresh on a scheduled cadence rather than per-run
      - Use Unify's pre-computed live benchmarks instead of custom runs where sufficient
costAllocation:
  tags:
    - name: endpoint
      description: model@provider routing target (e.g., gpt-4o@openai)
    - name: optimizer
      description: routing strategy used (quality, cost, latency, or custom)
    - name: project
      description: logical project grouping for multi-team cost attribution
    - name: tag
      description: arbitrary message tags for routing context and cost attribution
  description: >
    Unify supports message-level tagging via the API for cost attribution.
    Tags are arbitrary strings echoed back in responses, enabling downstream
    cost allocation by feature, team, or use case in the observability dashboard.
optimization:
  - name: Optimizer Routing
    description: >
      Use Unify's built-in optimizer strings (e.g., quality>0.8, cost, ttft)
      to automatically route each prompt to the cheapest provider meeting
      specified thresholds rather than hardcoding expensive flagship models.
  - name: Response Caching
    description: >
      Enable caching to serve repeated identical prompts from cache without
      incurring token costs. Particularly valuable for RAG pipelines, chatbot
      FAQ patterns, and development/testing loops.
  - name: Fallback Routing
    description: >
      Configure fallback provider chains so requests automatically retry on
      cheaper backup providers when a primary provider is unavailable or slow,
      avoiding wasted spend on timeouts.
  - name: BYOK (Bring Your Own Keys)
    description: >
      Supply your own provider API keys to bypass Unify's shared provider pools,
      using negotiated enterprise pricing or existing committed-use discounts with
      providers like OpenAI or Anthropic directly.
  - name: Credit Monitoring
    description: >
      Monitor credit consumption and set usage alerts via console.unify.ai to
      prevent unexpected spend exhaustion and maintain budget visibility.
observability:
  dashboard: https://console.unify.ai
  metrics:
    - Cost per request by model and provider
    - Token consumption (input vs output) by endpoint
    - Latency vs cost tradeoff visualization
    - Cache hit rate
    - Provider routing distribution
    - Credit burn rate over time
urls:
  - type: Console
    url: https://console.unify.ai
  - type: Pricing
    url: https://unify.ai/pricing
  - type: Documentation
    url: https://docs.unify.ai