@anura-gate/anura-graph

Typed TypeScript/JavaScript client for the Anura Memory API. Zero dependencies.

Anura Memory provides two memory products for AI agents:

• GraphRag — Knowledge graph with automatic triple extraction, deduplication, and hybrid retrieval
• FilesRag — Markdown file storage with heading-based chunking and semantic search

Installation

npm install @anura-gate/anura-graph

Quick Start

import { GraphMem } from '@anura-gate/anura-graph';

const mem = new GraphMem({
  apiKey: 'gm_your_key_here',
  // baseUrl: 'https://anuramemory.com' (default)
});

// --- GraphRag ---
await mem.remember("Alice is VP of Engineering at Acme Corp");
const ctx = await mem.getContext("Tell me about Alice");

// --- FilesRag ---
await mem.writeFile({ path: "/notes/standup.md", content: "# Standup\n## 2026-02-21\n- Shipped auth module" });
const results = await mem.searchFiles("auth module");

Configuration

new GraphMem({
  apiKey: 'gm_...',                    // required — from dashboard
  baseUrl: 'https://anuramemory.com',  // default
  retry: {
    maxRetries: 3,                     // default
    baseDelayMs: 500,                  // exponential backoff base
    maxDelayMs: 10_000,                // max delay cap
    retryOn: [429, 500, 502, 503, 504] // retryable status codes
  }
});

API Reference

GraphRag

remember(text): Promise<RememberResult>

Extract knowledge from text and store as triples in the graph.

const result = await mem.remember(
  "Albert Einstein was born in Ulm, Germany. He developed the theory of relativity."
);
// { extractedCount: 2, mergedCount: 2, conflicts: [], message: "..." }

If a new fact contradicts an existing one on a single-valued predicate (e.g., LIVES_IN, WORKS_AT), the old value is replaced and the conflict is returned:

const r = await mem.remember("Alice now works at Microsoft");
// r.conflicts → [{ subjectName: "alice", predicate: "WORKS_AT", oldObject: "google", newObject: "microsoft", resolution: "auto_recency" }]

getContext(query, options?): Promise<ContextResult>

Retrieve context from the knowledge graph using hybrid search (graph traversal + vector similarity + community summaries).

// Markdown format (ideal for LLM system prompts)
const ctx = await mem.getContext("Einstein", { format: "markdown" });
// ctx.content → "- Albert Einstein → BORN_IN → Ulm..."

// JSON format
const json = await mem.getContext("Einstein", { format: "json" });
// json.entities, json.edges, json.vectorEntities, json.communities

// Hybrid mode (graph + vector + communities)
const hybrid = await mem.getContext("Einstein", { mode: "hybrid" });

search(entity, options?): Promise<SearchResult>

Find an entity and its direct (1-hop) connections. In fuzzy mode, filter results by entity type.

const result = await mem.search("typescript");
console.log(result.entity); // { name: "Typescript", type: "technology", userId: "..." }
console.log(result.edges);  // all incoming + outgoing relationships

// Search with type filter (fuzzy mode)
const result = await mem.search("alice", { type: "person" });

ingestTriples(triples): Promise<IngestResult>

Ingest pre-formatted triples directly (no LLM extraction, max 100 per request).

const result = await mem.ingestTriples([
  { subject: "TypeScript", predicate: "CREATED_BY", object: "Microsoft" },
  { subject: "TypeScript", predicate: "SUPERSET_OF", object: "JavaScript" },
]);

getGraph(options?): Promise<GraphData>

Get the full graph (all nodes, edges, communities). Optionally filter by entity type.

const graph = await mem.getGraph();
// graph.nodes: GraphNode[], graph.edges: GraphEdge[], graph.communities?: GraphCommunity[]

// Filter graph by entity type
const people = await mem.getGraph({ type: "person" });
const orgs = await mem.getGraph({ type: "person,organization" });

deleteEdge(id, options?): Promise<void>

Delete an edge. Optionally blacklist the triple to prevent re-creation.

await mem.deleteEdge("edge_id");
await mem.deleteEdge("edge_id", { blacklist: true });

updateEdgeWeight(id, options?): Promise<void>

Set or increment an edge's weight. Higher-weight edges appear first in context retrieval.

await mem.updateEdgeWeight("edge_id", { weight: 5.0 });     // absolute
await mem.updateEdgeWeight("edge_id", { increment: 1.0 });  // delta

deleteNode(id): Promise<void>

Delete a node and all its connected edges.

exportGraph(): Promise<ExportData>

Export the entire graph as portable JSON.

importGraph(data): Promise<ImportResult>

Import a graph export into the current project (merges, does not delete existing data).

listCommunities(): Promise<Community[]>

List all detected communities.

detectCommunities(): Promise<DetectCommunitiesResult>

Run Louvain community detection + LLM summarization.

const result = await mem.detectCommunities();
// { communityCount: 3, entityCount: 25, message: "..." }

FilesRag

writeFile(options): Promise<WriteFileResult>

Create or update a markdown memory file. Files are automatically chunked by ## headings and indexed for semantic search.

const result = await mem.writeFile({
  path: "/docs/architecture.md",
  content: "# Architecture\n\n## Backend\nNode.js with Prisma...\n\n## Frontend\nNext.js with React...",
  name: "architecture.md", // optional, derived from path if omitted
});
// { file: { id, name, path, size, ... }, chunkCount: 3, created: true }

If a file already exists at the given path, its content is replaced and re-indexed.

listFiles(): Promise<MemoryFile[]>

List all files in the current project.

const files = await mem.listFiles();
// [{ id, name, path, size, createdAt, updatedAt }, ...]

readFile(id): Promise<FileWithContent>

Get a file with its content.

const file = await mem.readFile("file_id");
console.log(file.content); // full markdown content

updateFile(id, content, name?): Promise<WriteFileResult>

Update a file's content (re-chunks and re-indexes).

const result = await mem.updateFile("file_id", "# Updated content\n...");

deleteMemoryFile(id): Promise<void>

Delete a file and all its indexed chunks.

searchFiles(query, options?): Promise<FileSearchResult[]>

Semantic search across file chunks. Pass fileId to scope the search to a single file.

const results = await mem.searchFiles("authentication flow", { limit: 5 });
// Search within a single file:
// const results = await mem.searchFiles("auth", { fileId: "file_abc123" });
for (const r of results) {
  console.log(r.file.path);
  for (const chunk of r.chunks) {
    console.log(`  ${chunk.headingTitle} (${chunk.score.toFixed(2)})`);
    console.log(`  ${chunk.excerpt}`);
  }
}

Projects

listProjects(): Promise<Project[]>

createProject(name): Promise<Project>

deleteProject(id): Promise<void>

selectProject(id): Promise<void>

Traces

listTraces(options?): Promise<Trace[]>

getTrace(id): Promise<TraceDetail>

Blacklist

listBlacklist(options?): Promise<BlacklistedTriple[]>

addToBlacklist(subject, predicate, object): Promise<{ id: string }>

removeFromBlacklist(id): Promise<void>

Conflict Log

listConflicts(options?): Promise<ConflictLogEntry[]>

List conflict resolution log entries (newest first, paginated).

const conflicts = await mem.listConflicts({ limit: 10 });
// [{ id, subjectName, predicate, oldObject, newObject, resolution, createdAt }, ...]

Pending Facts

listPending(options?): Promise<PendingFact[]>

approveFact(id): Promise<void>

rejectFact(id, options?): Promise<void>

approveAll(): Promise<void>

rejectAll(): Promise<void>

Usage

getUsage(): Promise<UsageInfo>

const usage = await mem.getUsage();
// { tier: "FREE", factLimit: 100, currentFacts: 42, nodesCount: 20, edgesCount: 22,
//   fileStorageLimit: 52428800, fileCountLimit: 20, currentFileStorage: 1024, currentFileCount: 3 }

Health

health(): Promise<HealthResult>

Error Handling

All methods throw GraphMemError on failure:

import { GraphMemError } from '@anura-gate/anura-graph';

try {
  await mem.readFile("nonexistent");
} catch (err) {
  if (err instanceof GraphMemError) {
    console.log(err.status);  // 404
    console.log(err.message); // "File not found"
    console.log(err.body);    // full response body
  }
}

Rate Limiting

After each request, rate limit info is available:

await mem.remember("some fact");
console.log(mem.rateLimit.remaining); // requests remaining
console.log(mem.rateLimit.limit);     // total allowed per window
console.log(mem.rateLimit.reset);     // unix timestamp when window resets

Types

All types are exported:

import type {
  // GraphRag
  RememberResult, ConflictResolution, ConflictLogEntry,
  ContextResult, SearchResult, Triple,
  GraphNode, GraphEdge, GraphData, Community,
  // FilesRag
  MemoryFile, FileWithContent, FileSearchResult,
  WriteFileOptions, WriteFileResult,
  // Config
  GraphMemConfig, RetryConfig, RateLimitInfo, UsageInfo,
} from '@anura-gate/anura-graph';

License

MIT