AGPL-3.0-or-later · forever.

the provider catalog

Providers.

One provider/model line moves a workflow between models: local first, keys via env vars that stay yours, no inference lock-in. The same file runs on a laptop Ollama and a frontier API; swapping is a one-word diff. Every provider owns a room: the id in each row is its door. Machines read the catalog at /providers/catalog.json; the binary answers nika catalog.

  • 17named providersone provider: line
  • 5localthe sovereign default
  • 11cloudyour keys, your bill
  • 21embedded tailopenai-compatible

29 models the catalog knows, from 32k to 1049k tokens of context. 7 of them want no key at all.

  • mistral-small-latest32k
  • deepseek-chat64k
  • qwen3.5:4b128k
  • llama3.2:3b128k
  • qwen3.5-4b128k
  • llamacpp/default128k
  • localai/default128k
  • Qwen/Qwen3-8B128k
  • mistral-large-latest128k
  • llama-3.3-70b-versatile128k
  • llama-3.1-8b-instant128k
  • anthropic/claude-sonnet-4-20250514128k
  • anthropic/claude-haiku-4-5-20251001128k
  • meta/llama-3.1-70b-instruct128k
  • kimi-k2128k
  • grok-3131k
  • grok-3-mini-fast131k
  • claude-sonnet-4-20250514200k
  • claude-haiku-4-5-20251001200k
  • kimi-k2.5256k
  • Qwen/Qwen3.5-397B-A17B:fastest262k
  • Qwen/Qwen3.5-9B:cheapest262k
  • nvidia/nemotron-3-super-120b-a12b262k
  • nvidia/nemotron-3-nano-30b-a3b262k
  • gpt-5.2272k
  • gpt-5-mini272k
  • mock-default1000k
  • gemini-2.5-flash1049k
  • gemini-2.0-flash1049k

○ no key needed · ● a key, held by you · ◇ declared reasoning · ◆ takes images. A log scale, because the range is five doublings wide. Context and capabilities derive from the catalog the engine ships; the pricing rows it carries are empty today, so this page does not draw a price it does not have.

local5 providers

the sovereign default: your hardware, no key, nothing leaves the machine

  1. ollama#Ollamano key

    Local Ollama server (default port 11434) — serves any pulled model.

    default
    qwen3.5:4b
    cheap
    llama3.2:3b
    dialect
    openai-chat
    • qwenqwen3.5:4bctx 128k · vision
    • llamallama3.2:3bctx 128k · vision
  2. lmstudio#LM Studiono key

    LM Studio local server (default port 1234) — serves the loaded model.

    default
    qwen3.5-4b
    dialect
    openai-chat
    alias
    lm-studio
    • qwenqwen3.5-4bctx 128k · vision
  3. llamacpp#llama.cppno key

    llama.cpp server (default port 8080) — serves the loaded GGUF; the wire model id is free-form.

    default
    default
    dialect
    openai-chat
    alias
    llama-cpp
    • defaultdefaultctx 128k · vision
  4. localai#LocalAIno key

    LocalAI server (default port 8080) — an OpenAI-compatible gateway over local backends.

    default
    default
    dialect
    openai-chat
    • defaultdefaultctx 128k · vision
  5. vllm#vLLMno key

    vLLM server (default port 8000) — serves Hugging Face model ids at throughput.

    default
    Qwen/Qwen3-8B
    dialect
    openai-chat
    • qwenQwen/Qwen3-8Bctx 128k · vision
cloud11 providers

bring your own key; it rides an env var, never a config file, and stays yours

  1. mistral#Mistral AIMISTRAL_API_KEY

    Mistral Large, Medium, Small models.

    default
    mistral-large-latest
    cheap
    mistral-small-latest
    dialect
    openai-chat
    • largemistral-large-latestctx 128k · vision · json:schema
    • smallmistral-small-latestctx 32k · vision · json:schema
  2. anthropic#Anthropic ClaudeANTHROPIC_API_KEY

    Claude models (Opus, Sonnet, Haiku).

    default
    claude-sonnet-4-20250514
    cheap
    claude-haiku-4-5-20251001
    dialect
    anthropic
    alias
    claude
    • sonnetclaude-sonnet-4-20250514ctx 200k · reasoning · vision · json:schema
    • haikuclaude-haiku-4-5-20251001ctx 200k · reasoning · vision · json:schema
  3. openai#OpenAIOPENAI_API_KEY

    GPT-5.x, the o-series, and other OpenAI models.

    default
    gpt-5.2
    cheap
    gpt-5-mini
    dialect
    openai-chat
    alias
    gpt
    • gpt-5.2gpt-5.2ctx 272k · reasoning · vision · json:schema
    • gpt-5-minigpt-5-minictx 272k · reasoning · vision · json:schema
  4. gemini#Google GeminiGEMINI_API_KEY

    Gemini Pro, Flash, and Ultra models.

    default
    gemini-2.5-flash
    cheap
    gemini-2.0-flash
    dialect
    gemini
    alias
    google
    • flash-25gemini-2.5-flashctx 1.0m · vision · json:schema
    • flash-20gemini-2.0-flashctx 1.0m · vision
  5. deepseek#DeepSeekDEEPSEEK_API_KEY

    DeepSeek Chat and Coder models.

    default
    deepseek-chat
    dialect
    openai-chat
    alias
    deep-seek
    • chatdeepseek-chatctx 64k · json:object
  6. xai#xAI GrokXAI_API_KEY

    Grok models (Grok-3, Grok-4).

    default
    grok-3
    cheap
    grok-3-mini-fast
    dialect
    openai-chat
    alias
    grok
    • grok-3grok-3ctx 131k · json:schema
    • minigrok-3-mini-fastctx 131k · reasoning · json:schema
  7. groq#GroqGROQ_API_KEY

    Fast inference with Llama, Mixtral models.

    default
    llama-3.3-70b-versatile
    cheap
    llama-3.1-8b-instant
    dialect
    openai-chat
    • llama-70bllama-3.3-70b-versatilectx 128k · json:schema
    • llama-8bllama-3.1-8b-instantctx 128k · json:schema
  8. openrouter#OpenRouterOPENROUTER_API_KEY

    200+ models via unified gateway.

    default
    anthropic/claude-sonnet-4-20250514
    cheap
    anthropic/claude-haiku-4-5-20251001
    dialect
    openai-chat
    alias
    or
    • defaultanthropic/claude-sonnet-4-20250514ctx 128k · reasoning · vision · json:schema
    • cheapanthropic/claude-haiku-4-5-20251001ctx 128k · reasoning · vision · json:schema
  9. huggingface#Hugging FaceHF_TOKEN

    Inference Providers router · 100+ open-weight models across 18 providers (Groq · Cerebras · Together · Scaleway · OVHcloud · …) · zero markup · :provider or :fastest/:cheapest routing suffix.

    default
    Qwen/Qwen3.5-397B-A17B:fastest
    cheap
    Qwen/Qwen3.5-9B:cheapest
    dialect
    openai-chat
    alias
    hf
    • defaultQwen/Qwen3.5-397B-A17B:fastestctx 262k · vision
    • cheapQwen/Qwen3.5-9B:cheapestctx 262k · vision
  10. nvidia#NVIDIANVIDIA_API_KEY

    NVIDIA API (integrate.api.nvidia.com) · Nemotron 3 family (Open Model License · agentic-first) + hosted open models · self-hosted NIM containers expose the same surface.

    default
    nvidia/nemotron-3-super-120b-a12b
    cheap
    nvidia/nemotron-3-nano-30b-a3b
    dialect
    openai-chat
    alias
    nim · nvidia-nim
    • defaultnvidia/nemotron-3-super-120b-a12bctx 262k · json:schema
    • cheapnvidia/nemotron-3-nano-30b-a3bctx 262k · json:schema
    • llama-70bmeta/llama-3.1-70b-instructctx 128k · json:schema
  11. moonshot#Moonshot (Kimi)MOONSHOT_API_KEY

    Moonshot AI — Kimi K2, K2.5 (international endpoint api.moonshot.ai/v1). rig-core native provider.

    default
    kimi-k2
    dialect
    openai-chat
    alias
    kimi
    • k2kimi-k2ctx 128k · json:schema
    • k2-5kimi-k2.5ctx 256k · json:schema
test1 provider

the harness: mock-first is how workflows are written before they cost anything

  1. mock#Mockno key

    Deterministic test responses — no API calls, no keys needed.

    default
    mock-default
    dialect
    mock
    • defaultmock-defaultctx 1m · reasoning · vision · json:schema

Beyond the named set, the engine embeds 21 more OpenAI-compatible endpoints; nika catalog lists every one. Models are aliases in the file (model: sonnet) pinned to exact upstream ids by the engine, so a run is reproducible even after a vendor rotates names. Try one in the playground, or install and go local first. Read the spec →