Developers

A decision in front of your memory store.

Bring your own model and store. Gromani decides what to retain, update, forget or escalate, and keeps the record. The surface below is illustrative and will change.

import { Gromani } from "@gromani/sdk";

const memory = new Gromani({ policy: "eng-default@3" });

const result = await memory.propose({
  agentId: "release-bot",
  subject: "release:atlas-2.4",
  statement: "Release moves to 11 Dec after the load-test regression.",
  source: { channel: "standup-notes", authority: "owner" },
});

result.decision; // "update"
result.rule;     // "R5"
result.audit.id; // "aud_1h7x2k9q"

propose: Submit a candidate memory and get a decision, a rule and an audit id.
recall: Read memories for a subject, filtered by privacy class and confidence.
forget: Remove a memory on request or on schedule; the tombstone stays.
audit.list: Walk the decision trail for a subject, an agent or a time window.
review.queue: Everything escalated and waiting for a person, with context.

Your choice of model

Designed to work with the models you choose. Tested models, versions and limits will be published.

Store-agnostic

Works in front of a key-value store, a vector index or a graph. Gromani holds the decision record; you hold the memories.

Audit first

Every write, change and removal is an entry you can list, filter and export. The policy version is part of the entry.

The SDK is not yet published. Join the waitlist to hear when early access opens.