Intelital

For builders

Judgment,
over an API.

Your agents can act. Give them a brain to consult first: ranked, explained decisions from one call.

Read the docsBook a demo →
api.intelital.com
GET /v1/decisions?subject=SKU-4471

{
  "decision": "reorder_now",
  "confidence": 0.84,
  "subject": "SKU-4471",
  "rank": { "position": 2, "of": 1840 },
  "traced_from": [
    { "type": "signal",  "fact": "lead_time_slipped_12d_to_19d" },
    { "type": "pattern", "fact": "matches_2_prior_stockouts" },
    { "type": "rank",    "fact": "#2_stockout_risk_of_1840" }
  ],
  "expires_at": "2026-06-11T09:00:00Z"
}

The loop

Three endpoints. A learning loop.

Ask for a decision. Report what happened. The model re-ranks from the outcome. That correction sharpens the next ranking.

GET/v1/decisions

Ask for a ranked decision. You get the answer, confidence, rank position, and what it traced from.

200 · response
{
  "decision": "reorder_now",
  "confidence": 0.84,
  "rank": { "position": 2, "of": 1840 }
}

POST/v1/outcomes

Tell it what happened. This is how it gets smarter.

request
POST /v1/outcomes
{
  "decision_id": "dec_8f3a",
  "result": "no_stockout",
  "acted_at": "2026-06-11T14:20:00Z"
}

GET/v1/decisions/{id}/trace

The receipts, machine-readable. Your agent can explain itself.

200 · response
{
  "traced_from": [
    {
      "type": "signal",
      "fact": "lead_time_slipped_12d_to_19d",
      "source": "carrier_edi"
    }
  ]
}

The loop is the product. Decisions without outcome reporting plateau; the feedback is where the compounding lives.

Integration patterns

Three ways agents use it.

01

The before-act check.

Agent consults Intelital before any expensive action. It acts only above a confidence threshold, and escalates to a human below it.

before_act.py
r = requests.get(
  "https://api.intelital.com/v1/decisions",
  params={"subject": "SKU-4471"},
  headers=AUTH,
).json()

if r["confidence"] >= 0.80:
  place_order(r["subject"])  # act
else:
  escalate(r, to="ops-oncall")  # ask a human first

02

Queue consumption.

Pull today's ranked decisions, work them top down, and report outcomes back as they land. The queue re-ranks underneath you.

consume_queue.py
queue = requests.get(
  "https://api.intelital.com/v1/decisions",
  params={"rank": "top", "limit": 20},
  headers=AUTH,
).json()

for decision in queue["items"]:
  result = process(decision)
  report_outcome(decision["id"], result)

03

Human-in-the-loop override.

Human overrides a decision via webhook. The override is a label, not an exception — it trains the next ranking directly.

webhook.py
@app.post("/webhooks/intelital")
def on_override(event):
  o = event["override"]
  # an override is a label, not an exception
  label(o["decision_id"], o["chosen_action"])
  return 200
MCP Server

Plug judgment into Claude or any agent framework as a native tool.

mcp.json
{ "mcpServers": { "intelital": {
  "url": "https://mcp.intelital.com",
  "auth": "bearer"
} } }

The practicals

142msDecision endpoint, p95 latency
BearerAPI-key auth, Authorization header
IdempotentOutcome reporting, safe to retry
IncludedSandbox demo workspace, every key

Every key ships with the demo workspace: 4,206 entities, live decisions, no setup. Your first call returns a real decision in under five minutes.

Quickstart

Five minutes to your first decision.

01

Get a key

Grab a sandbox key from the dashboard and export it. The demo workspace is already wired up.

shell
export INTELITAL_KEY="sk_live_xxxxxxxxxxxxxxxx"
02

Ask for a decision

Curl the decisions endpoint with the sandbox subject. You get a ranked, explained decision back.

shell
curl "https://api.intelital.com/v1/decisions?subject=SKU-4471" \
-H "Authorization: Bearer $INTELITAL_KEY"
03

Open the trace

Follow the decision id into its trace to see the signals, patterns, and rankings it reasoned from.

shell
curl "https://api.intelital.com/v1/decisions/dec_8f3a/trace" \
-H "Authorization: Bearer $INTELITAL_KEY"
Full reference docs →

Give your agents judgment.