schema_version: "1.0"

product:
  name: Bifrost
  type: ai_execution_layer
  description: "Intelligent AI execution layer that routes, optimizes, caches, validates, and recovers AI requests across multiple model providers"
  website: https://bifrost.omixsystems.store
  version: "0.1.0"
  author: OMIX Digital Solutions
  location: Kericho, Kenya

api:
  type: openai_compatible
  base_url: https://bifrost.omixsystems.store/v1
  auth: bearer_token
  endpoints:
    - method: POST
      path: /v1/chat/completions
      description: Chat completions with intelligent routing and optimization
    - method: GET
      path: /v1/models
      description: List all available models across providers

capabilities:
  routing:
    enabled: true
    description: "Automatic model/provider selection based on capability, cost, latency, health, and quota"
    scoring:
      capability_match: 0.30
      quality: 0.25
      reliability: 0.15
      cost_efficiency: 0.15
      latency: 0.10
      availability: 0.05
  optimization:
    enabled: true
    description: "Prompt compression, tool output compression, context pruning, deduplication"
    features:
      - prompt_compression
      - tool_output_compression
      - semantic_cache
      - recoverable_context
      - context_dependency_graph
      - context_garbage_collection
  caching:
    enabled: true
    types:
      - exact_match
      - semantic_similarity
  reliability:
    enabled: true
    features:
      - automatic_fallback
      - circuit_breakers
      - provider_cooldowns
      - self_healing
      - backpressure
      - priority_queues
      - stream_keepalive
      - multi_account_rotation
  control:
    enabled: true
    features:
      - policy_as_code
      - budgets
      - quotas
      - data_residency
      - tenant_isolation
      - provider_restrictions
  observability:
    enabled: true
    features:
      - request_tracing
      - cost_attribution
      - provider_health
      - routing_decisions
      - fallback_events

providers:
  - id: ollama
    name: Ollama
    type: local
    models: [llama3, mistral, llama3.1, gemma2, codellama, deepseek-r1, qwen2.5]
  - id: openai
    name: OpenAI
    type: paid
    models: [gpt-4o, gpt-4o-mini, gpt-4-turbo]
  - id: gemini
    name: Google Gemini
    type: free_tier
    models: [gemini-2.0-flash, gemini-2.5-pro, gemini-2.5-flash]
  - id: groq
    name: Groq
    type: free_tier
    models: [llama-3.3-70b-versatile, mixtral-8x7b-32768]
  - id: cerebras
    name: Cerebras
    type: free_tier
    models: [llama-3.3-70b, llama-3.1-8b]
  - id: sambanova
    name: SambaNova
    type: free_tier
    models: [Meta-Llama-3.3-70B-Instruct, DeepSeek-R1]
  - id: openrouter
    name: OpenRouter
    type: free_tier
    models: [meta-llama/llama-3.3-70b-instruct:free, qwen/qwen-2.5-72b-instruct:free]
  - id: cloudflare
    name: Cloudflare Workers AI
    type: free_tier
    models: [@cf/meta/llama-3.3-70b-instruct-fp16]
  - id: mistral
    name: Mistral AI
    type: free_tier
    models: [mistral-small-latest, mistral-large-latest]
  - id: huggingface
    name: HuggingFace
    type: free_tier
    models: [meta-llama/Llama-3.3-70B-Instruct]
  - id: ollama-cloud
    name: Ollama Cloud
    type: free_tier
    models: [gemma4:31b, gpt-oss:20b, nemotron-3-nano:30b]
  - id: zen
    name: Zen
    type: paid
    models: [zen-lite, zen-pro]
  - id: bytez
    name: Bytez
    type: paid
    models: [bytez-pro]

total_models: 48
total_providers: 14