The mixture of jevs,
as a model.

One POST /v1/systemone fans your question out to several independent decision-model families in parallel, pools their probability distributions, and answers in under a second — in Jev's own dialect, so any Jev-compatible client works by switching the base URL. Every fresh decision carries a public receipt. Every repeat is free.

Quickstart

Self-serve signup mints a free-tier key instantly (shown once — store it):

# 1 · get a key (free tier: 1,000 decisions/mo)
curl -s https://jevcache.sh/api/v1/orgs/signup \
  -H 'content-type: application/json' \
  -d '{"email":"you@example.com","use_case":"routing support tickets"}'
# → { "key": "sk_hs_…", "tier": "free", … }

# 2 · ask the mixture
curl -s https://jevcache.sh/api/v1/systemone \
  -H "authorization: Bearer $KEY" -H 'content-type: application/json' \
  -d '{
    "model": "hyperswarm/swarm-1",
    "state": { "ticket": "Customer says the export button 500s on files over 2GB." },
    "questions": {
      "severity": { "type": "score", "instructions": "How severe is this bug?",
                    "criteria": ["cosmetic", "minor", "major", "critical"] },
      "refund":   { "type": "noul",  "instructions": "Should support offer a refund now?" }
    }
  }'

A typical response:

{
  "model": "hyperswarm/swarm-1",
  "answers": {
    "severity": { "type":"score", "score":2, "confidence":0.84,
                  "legend":{"0":"cosmetic","1":"minor","2":"major","3":"critical"},
                  "probabilities":{"0":0.01,"1":0.09,"2":0.84,"3":0.06} },
    "refund":   { "type":"noul", "noul":0.22, "confidence":0.56 }
  },
  "from_memory": false,
  "latency_ms": 612,
  "voices": [ {"family":"pplx","model":"pplx-decider-v1-27b","ms":540},
              {"family":"clef","model":"@cf/cloudflare/clef-flash","ms":480} ],
  "receipt": "https://jevcache.sh/d/q_8208…",
  "question_id": "q_8208…",
  "billing": { "model":"hyperswarm/swarm-1", "charged_usd":0.001, "tier":"free" }
}

Ask the identical question again and from_memory: true comes back in ~0.35s, billed at $0.000001 — the commons remembers.

Models

Model sets are the price/quality dial. Quality comes from diversity: independent families rarely share failure modes, and the pooled posterior is measurably better calibrated than any single voice. Live catalog: GET /v1/models (public).

Pricing

Bring your own keys

The goal is the commons: decisions computed once, remembered forever, served free. If you'd rather pay your model providers than us, send model: "hyperswarm/swarm-byo-1" with your keys as headers — we orchestrate the quorum, pool the posterior, seal the receipt, and charge nothing:

curl -s https://jevcache.sh/api/v1/systemone \
  -H "authorization: Bearer $KEY" \
  -H "x-byo-jev: $TYPESAFE_KEY" \
  -H "x-byo-pplx: $PPLX_KEY" \
  -H "x-byo-liquid: $LIQUID_KEY" \
  -d '{ "model": "hyperswarm/swarm-byo-1", "state": …, "questions": … }'

Keys live in request memory only — never stored, never logged, never in usage events (x-byo-clef + x-byo-clef-account for Cloudflare). On the house models the same headers substitute your key for ours per family. Every fresh BYO decision seals into the public commons, where anyone's identical ask replays it free — that is the point.

To be clear about what the fee buys: the underlying decision models are token-priced and tiny — a single jev call is thousandths of a cent ($0.042/M input, output free). swarm-1's $0.001 is an orchestration fee: the parallel quorum across four independent families, the pooled calibrated posterior, a permanent public receipt, and the commons write. It is not a cheaper jev; it is a better-calibrated answer with an audit trail. And you are never billed for a decision the commons already holds.

Tiers

Upgrade path: POST /v1/waitlist with your email and expected volume. Check your live usage anytime with GET /v1/org; the full event log is GET /v1/usage.

The dialect

Requests and answers speak Jev's decisions dialect. Three question types:

state is any JSON — the facts the question is about. The fingerprint (and so the memory) is computed over (schema, state), which is why identical asks are free.

Model directives

Suffixes on the model string tune one call without new parameters:

"model": "hyperswarm/swarm-1:fresh:a1.3"

Receipts

Every fresh decision opens a public arena question; the family attestations are written back after your response returns, the quorum seals, and the receipt at https://jevcache.sh/d/<question_id> shows the question, each family's vote, and the sealed consensus — a permanent, linkable audit trail. The hourly ledger digest is anchored to the A1 chain.

Idempotency

Send an Idempotency-Key header on any /v1/systemone or /v1/decide POST. Retries with the same key replay the stored response (marked x-idempotent-replay: 1) and are never re-billed.

OpenAI-compatible door

Any chat-completions client can use the swarm by switching the base URL and model string — POST /v1/chat/completions with model: "hyperswarm/swarm-1". Plain prose in the last user message becomes a yes/no oracle; a JSON body with {state, questions} runs verbatim. The receipt rides in response.hyperswarm.

# python / openai sdk
client = OpenAI(base_url="https://jevcache.sh/api/v1", api_key=KEY)
r = client.chat.completions.create(
    model="hyperswarm/swarm-1",
    messages=[{"role":"user","content":"Will this PR description pass review as written?"}])

Errors

Error bodies are always {"error": {"type", "message"}}.

Access & status

GET /v1/status live arena + pool health, p50 · GET /v1/models catalog + prices + families up · GET /v1/health storage liveness for monitors — all public, all edge-cached.

Key hygiene: keys are shown once and stored only as hashes. Rotate anytime with POST /v1/keys/rotate — the old key dies instantly.

© 2026 HyperspaceAI™ · terms · privacy · jevcache · the decision commons — compute the decision once, remember it forever. Watch the swarm work at hyperswarm.sh.