Drop-in trust layer.
The same code you already write.

using Vorana.Client;
using Vorana.Shared;

// In Program.cs / Startup
services.AddVoranaClient(opts =>
{
    opts.BaseAddress       = new Uri("http://localhost:5057");
    opts.ApiKey            = "vorana-dev-key";
    opts.DefaultPipelineId = "pipeline.faq_assistant.v1";
});

// Anywhere IVoranaClient is injected
var resp = await client.GenerateAsync(new GenerateRequest
{
    Inputs = new() { ["user_query"] = "What is our refund policy?" }
});

Console.WriteLine(resp.Output["content"]);
Console.WriteLine($"run_id={resp.RunId}, confidence={resp.Confidence}");

const res = await fetch('https://gateway.vorana.ai/v1/generate', {
  method:  'POST',
  headers: {
    'X-Api-Key':    process.env.VORANA_API_KEY!,
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    pipeline_id: 'pipeline.faq_assistant.v1',
    inputs:      { user_query: "What is our refund policy?" },
  }),
});

const data = await res.json();
console.log(data.output.content, data.confidence, data.run_id);

import httpx, os

resp = httpx.post(
    "https://gateway.vorana.ai/v1/generate",
    headers={"X-Api-Key": os.environ["VORANA_API_KEY"]},
    json={
        "pipeline_id": "pipeline.faq_assistant.v1",
        "inputs":      {"user_query": "What is our refund policy?"},
    },
).json()

print(resp["output"]["content"], resp["confidence"], resp["run_id"])

{
  "run_id":    "r_8f3a2c…",
  "trace_id":  "01J5HQ…",
  "status":    "ok",                          // ok | low_confidence | denied | failed
  "output": {
    "content":    "Refunds within 30 days, full…",
    "confidence": 0.92
  },
  "confidence": 0.92,
  "decisions": [
    { "step": "validation.input", "rule_or_policy": "schema",    "result": "pass" },
    { "step": "llm.consensus",    "rule_or_policy": ">= 0.85",   "result": "pass", "reason": "0.92" },
    { "step": "scoring.composite","rule_or_policy": "weighted",  "result": "pass" }
  ],
  "citations": [
    { "source": "policy/refunds.pdf", "section": "§3.2" }
  ],
  "metrics": {
    "latency_ms":        1500,
    "prompt_tokens":     280,
    "completion_tokens": 120,
    "cost_usd":          0.0142
  },
  "obligations": []
}

// Browser SSE — Vorana also accepts query-string auth for EventSource
const es = new EventSource(
  '/v1/generate:stream?api_key=vorana-dev-key' +
  '&pipeline_id=pipeline.faq_assistant.v1' +
  '&inputs.user_query=Refund%20window%3F'
);

es.onmessage = (e) => {
  const chunk = JSON.parse(e.data);
  if (chunk.event === 'token')  appendToUi(chunk.content);
  if (chunk.event === 'done')   { saveRunId(chunk.run_id); es.close(); }
};

import httpx, json, os

with httpx.stream(
    "POST", "https://gateway.vorana.ai/v1/generate:stream",
    headers={"X-Api-Key": os.environ["VORANA_API_KEY"]},
    json={"pipeline_id": "pipeline.faq_assistant.v1",
          "inputs":      {"user_query": "Refund window?"}},
) as r:
    for line in r.iter_lines():
        if not line.startswith("data:"): continue
        chunk = json.loads(line[5:])
        if chunk["event"] == "token":
            print(chunk["content"], end="", flush=True)
        elif chunk["event"] == "done":
            print(f"\nrun_id={chunk['run_id']}")

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://gateway.vorana.ai/v1',
  apiKey:  process.env.VORANA_API_KEY!,
});

const stream = await client.chat.completions.create({
  model:    'vorana:pipeline.faq_assistant.v1',
  messages: [{ role: 'user', content: "What's our refund policy?" }],
  stream:   true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? '');
}

curl -N https://gateway.vorana.ai/v1/chat/completions \
  -H "Authorization: Bearer vorana-dev-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "vorana:pipeline.faq_assistant.v1",
    "messages": [{ "role": "user", "content": "Refund policy?" }],
    "stream": true
  }'

# From your skill repo — package, sign, push to the org registry
$ vorana skill build   ./redact_customer_pii
→ bundle: dist/redact_customer_pii-2.1.0.tgz   (12 KiB)

$ vorana skill sign    ./dist/redact_customer_pii-2.1.0.tgz \
    --key acme-root.pem
→ signature OK

$ vorana skill publish ./dist/redact_customer_pii-2.1.0.tgz \
    --org acme
→ published skill.acme.security.redact_customer_pii@2.1.0
→ downstream pipelines: 14  (use --check to preview impact)

# Reference a published skill from any pipeline in your org
id:       pipeline.support_copilot.v3
steps:
  - id:   redact
    type: skill.acme.security.redact_customer_pii  # <-- here
    version: 2.1.0                                # pinned (recommended for prod)

  - id:   answer
    type: llm.execute
    with: { provider: azure_openai, model: gpt-4o }

  - id:   judge
    type: skill.acme.legal.judge_compliance         # another shared skill
    version: "~1.4"                              # floating: auto-track patches

# Inspect what's available in your org and where each skill is used
$ vorana skill list --org acme
NAME                                  VERSION  OWNER             USED BY
redact_customer_pii                   2.1.0    @acme/security    14 pipelines
judge_compliance                      1.4.2    @acme/legal       8  pipelines
retrieve_policy_kb                    3.0.1    @acme/platform    22 pipelines
summarize_call_note                   0.9.0    @acme/cx          6  pipelines

$ vorana skill describe redact_customer_pii
→ v2.1.0   signed by @acme/security   published 2026-04-02
→ used by pipelines:
    customer-chat-prod, support-copilot, claims-intake, advisor-agent,
    fraud-triage, broker-assistant, ... (10 more)

# Re-execute against today's pipeline + policy. Useful for regression tests,
# A/B'ing prompt changes, or verifying a fix to a low-confidence run.
curl -X POST -H "X-Api-Key: vorana-dev-key" \
  http://localhost:5057/v1/replay/r_8f3a2c…

# Optional body: { "pipeline_id": "...override...", "policy_id": "..." }

# Search by tenant, status, score, or text. Indexed via Postgres in production.
curl -H "X-Api-Key: vorana-dev-key" \
  "http://localhost:5057/v1/audit/search?tenant=tenant_acme&status=low_conf&from=2026-04-01&q=refund"

# →  { "runs": [ { "run_id": "...", "status": "low_conf", "confidence": 0.74, ... } ], "next": "..." }