java

Lians Java SDK

Financial-grade agent memory for the JVM — bitemporal recall, SEC 17a-4 audit chain, GDPR/HIPAA crypto-shred, information barriers, and a relationship graph for conflict-of-interest / related-party / care-network queries.

Built for where Java already runs: bank and insurer risk systems, healthcare platforms, and legal tech. Java 11+, one dependency (Jackson); HTTP is the JDK's java.net.http.

Install

Maven:

<dependency>
  <groupId>ai.lians</groupId>
  <artifactId>lians-sdk</artifactId>
  <version>0.2.0</version>
</dependency>

Gradle:

implementation "ai.lians:lians-sdk:0.2.0"

Quick start

import ai.lians.LiansClient;
import ai.lians.LiansClientOptions;
import ai.lians.model.MemoryOut;
import ai.lians.model.RecallResult;
import java.time.Instant;
import java.util.Map;

LiansClient client = new LiansClient(LiansClientOptions.builder()
        .baseUrl("https://api.lians.dev")              // or your self-hosted server
        .apiKey(System.getenv("LIANS_API_KEY"))
        .adminSecret(System.getenv("LIANS_ADMIN_SECRET"))   // only for /v1/admin/* audit calls
        .build());

// Store a fact with its BUSINESS event-time (not now)
client.addMemory("equity-desk", "NVDA FY2026 revenue guidance raised to $40B",
        Instant.parse("2025-11-19T16:00:00Z"),
        Map.of("ticker", "NVDA", "metric", "revenue_guidance"));

// Recall current (non-stale) facts
RecallResult r = client.recall("equity-desk", "NVDA revenue guidance", 5);
for (MemoryOut m : r.memories) {
    System.out.println(m.eventTime + "  " + m.content);
}

// Point-in-time — what did we know on a past date?
RecallResult past = client.recallAt("equity-desk", "NVDA revenue guidance",
        Instant.parse("2025-09-01T00:00:00Z"), 5);

Compliance & graph surfaces

// Exhaustive knowledge state at a date (regulator demo)
client.snapshot("equity-desk", Instant.parse("2026-03-01T00:00:00Z"), 1000);

// Lookahead-bias proof before trusting a backtest
client.backtestCheck("equity-desk", Instant.parse("2026-01-01T00:00:00Z"));

// GDPR/HIPAA crypto-shred + verify the tamper-evident chain
client.eraseSubject("MRN-00042", "GDPR-REQ-2026-001");
client.verifyChain("your-namespace");   // requires adminSecret

// Relationship graph — conflict-of-interest / related-party reachability
client.relate("matter-7", "Attorney", "represented", "ClientX",
        Instant.parse("2026-01-01T00:00:00Z"), false, false);
client.relate("matter-7", "ClientX", "adverse_to", "PartyY",
        Instant.parse("2026-01-01T00:00:00Z"), false, false);
var coi = client.path("matter-7", "Attorney", "PartyY", 4, null);
// -> {"connected": true, "hops": 2, "path": [...]}

// Graph-proximity reranking
client.recallNear("equity-desk", "earnings", "FundA", "ticker", 5);

Notes

• Timestamps are java.time.Instant (serialized as ISO-8601 UTC).
• Errors (non-2xx) throw LiansException exposing status() and body().
• Responses with rich schemas (snapshot, graph, conflicts, audit) return Jackson JsonNode; addMemory/recall return typed MemoryOut / RecallResult.
• LiansClient is thread-safe — create one and share it.

Why Java + Lians

mem0 ships Python/TypeScript only; Zep adds Go. Neither offers a Java SDK — yet Java is the backbone of regulated back-offices. This SDK brings the full compliance memory layer (and the relationship graph that neither competitor pairs with an open compliance spine) to the JVM. See the mem0 and Zep/Graphiti comparisons.

Build & test

cd agentmem/sdk/java
mvn test      # JUnit tests run against an in-process mock server — no live Lians needed
mvn package