the provider catalog
Groq
Custom LPU hardware serving open models at extreme tokens per second. Groq is the chip company, not the model. No relation to Grok. One provider/model line moves a workflow here or away again: the same file runs on every provider the catalog names, and swapping is a one-word diff. Machines read the catalog; the binary answers nika catalog.
- 2pinned modelsaliases pin exact upstream ids
- cloudfamilybring your own key; it rides an env var and stays yours
- GROQ_API_KEYthe key rides herean env var, never a config file
- openai-chatwire dialectthe adapter speaks it
the wire
cloud · bring your own key; it rides an env var and stays yoursA workflow names this provider in the model line (model: groq/llama-3.3-70b-versatile), and the engine speaks openai-chat on the wire. The key rides GROQ_API_KEY in the environment, never a config file, never the workflow.
- default
llama-3.3-70b-versatile- cheap
llama-3.1-8b-instant- dialect
openai-chat- tags
- fast · function-calling · parallel-tools · streaming · structured-output
the models
2 pinsAliases are what the file may write; the engine pins each one to an exact upstream id, so a run stays reproducible after a vendor rotates names.
llama-70bllama-3.3-70b-versatilectx 128k · json:schemallama-8bllama-3.1-8b-instantctx 128k · json:schema$0.05/M in · $0.08/M out
prices are the ENGINE's, not ours: the audit's pricing snapshot (models.dev · pinned 2026-07-07): the same table nika check prices your ceiling with.
in the released catalog
2 roomsThe wire models this provider serves in the released binary's catalog · each one is a room with its seats, exact prices and measured energy.
- llama-3.1-8b-instantas llama-8b
- llama-3.3-70b-versatileas llama-70b
the whole register: /catalog/models · what the released binary knows, vendored at the engine pin.
in a real file
a complete workflow · shown wholenika: v1workflow: id: instant-fanout description: "four prompts at once on LPU hardware — a failed item yields null, the batch lives"tasks: slogan: for_each: ["latency", "cost", "recovery", "provenance"] max_parallel: 4 on_error: recover: null infer: model: groq/llama-3.1-8b-instant max_tokens: 30 prompt: | A four-word slogan about ${{ item }} in CI pipelines.outputs: slogans: ${{ tasks.slogan.output }}check-green against the released engine · 1 task · 1 wave · worst-case ceiling $0.0000 · the drift gate re-proves this copy on every test run · open it in the playground →
get set up
keys at console.groq.com- 1Create a key in the console.
- 2The key rides the environment, never the file
export GROQ_API_KEY=…
Pick it for wide fan-outs and interactive loops where latency is the feature. Then prove the seat: nika doctor names every provider this binary can reach.
swap it
10 more cloud seatsMoving off Groq is editing one line. The rest of the file (tasks, permits, outputs, the trace) does not change. The register orders the catalog local first; beyond it the engine embeds 21 more OpenAI-compatible endpoints.
cross-references
the words that meet it
Try it in the playground, or install and go local first. Read the spec →