Use cases

Jepela is a hosted decision API with memory. You send a state and typed questions: a choice, a score, or a noul (the probability of yes). You get typed answers with probabilities in one call. Name a subject, and its memory is used too: lines in words, and rules checked in code. Jepela does not generate text, so it fits where your code needs a judgment it can branch on, instead of an LLM whose text you would have to parse.

Each page below shows a request copied from the docs, what was measured and when, and where the engine is weak.

Ticket routing and escalation

Pick the team, the urgency and whether a person must step in, for every ticket in one call. Write an account's standing facts to its memory once, and every later ticket from that account uses them. In the escalation cookbook (2026-09-23, on a test setup), two memory lines moved "escalate" for a priority account from 0.20 to 0.82, while a control account with no memory did not move. Read Ticket routing and escalation with memory.

Intent routing

One choice question over your intents, with a probability for every intent and a confidence to gate on. Up to 20 intents are read in one engine pass; above that, Jepela matches by default. On 600 real bank messages with 77 intents (2026-09-27, on a test server), matching was right on 65.2% of them against 62.2% for the engine reading all 77. Read Intent routing across many intents.

Decisions inside agents

Typed checks instead of free text: should this action run, which tool fits, does the user want a person. Jepela ships as an MCP server and a Claude Code plugin, keeps a memory per user, and removes options that must never be chosen in code. Read Typed decisions inside AI agents.

Screening and guardrails

Yes/no screening for spam, phishing, personal data and jailbreak attempts, with thresholds you set from your own error costs. In Jepela's measurements, spam and phishing were the strong case and moderation the weak one. Read Spam, phishing and guardrail screening.

Live examples

The demos call the real engine. Nothing is recorded or scripted: when the engine is off, they say so instead of answering.

  • The demo gallery: small apps that decide with memory, such as a calendar that keeps your rules, a support desk, a kitchen that remembers allergies, and forgetting on request.
  • Jepela Arcade: games in which two players make the same decisions with the same engine, and only one of them has a Jepela memory.

Price and accounts

$0.042 per million input tokens; output is free. Accounts are open: sign up, confirm your email address and pay by card to start at once (or wait for Jepela to approve the account), then create keys and add prepaid credit by card or invoice.

Next steps