@prometheus-ai/memory

Local SQLite memory engine for Prometheus agents.

This package is the Bun/TypeScript port of the Mnemosyne memory engine. It provides:

• PrometheusMemory, a small facade for remember/recall/stats/sleep workflows.
• BeamMemory, the lower-level working/episodic memory engine.
• MCP tool definitions and a dispatcher for host integrations.
• Optional local ONNX embeddings through fastembed and optional OpenAI-compatible embedding/LLM endpoints.

The package does not bundle or download a local GGUF LLM. LLM paths are host-backend or OpenAI-compatible remote only; when no LLM is configured, deterministic heuristic paths are used.

Basic use

import { PrometheusMemory } from "@prometheus-ai/memory";

const memory = new PrometheusMemory({ dbPath: "./prometheus-memory.db", bank: "project" });
const id = memory.remember("The deployment target is stable-cluster.", {
	source: "notes",
	importance: 0.8,
	veracity: "true",
});

const results = memory.recall("deployment target", 5);
console.log(id, results[0]?.content);

memory.close();

Configuration

PrometheusMemory accepts LLM and embedding options directly. PROMETHEUS_MEMORY_* environment variables remain fallbacks/defaults when the matching constructor option is omitted.

import { PrometheusMemory } from "@prometheus-ai/memory";
import type { Model } from "@prometheus-ai/ai";

const ftsOnly = new PrometheusMemory({ noEmbeddings: true });

const remoteEmbeddings = new PrometheusMemory({
	embeddingModel: "text-embedding-3-small",
	embeddingApiUrl: "https://api.openai.com/v1",
	embeddingApiKey: process.env.OPENAI_API_KEY,
});

const remoteLlm = new PrometheusMemory({
	llm: {
		baseUrl: "https://api.openai.com/v1",
		apiKey: process.env.OPENAI_API_KEY,
		model: "gpt-4.1-mini",
	},
	// Equivalent aliases: llmBaseUrl, llmApiKey, llmModel.
});

declare const smolModel: Model;
const modelBackedMemory = new PrometheusMemory({ llm: smolModel });
const dynamicLlm = new PrometheusMemory({
	llm: async (prompt, opts) => {
		const token = await getFreshOauthToken();
		return await completeWithModel(prompt, {
			token,
			maxTokens: opts?.maxTokens,
			temperature: opts?.temperature,
		});
	},
});

Banks and host scoping

PrometheusMemory itself exposes banks directly through constructor options such as bank; it does not hard-code coding-agent project scoping.

The Prometheus coding-agent wrapper adds prometheusMemory.scoping on top of those constructor options:

• global: one shared bank
• per-project: isolated project memory
• per-project-tagged: project-local writes plus global recall visibility

In per-project-tagged, the wrapper is responsible for combining project-local retention with global recall visibility. The package still just exposes banks plus constructor-level LLM and embedding options.

Common environment fallbacks:

• PROMETHEUS_MEMORY_DATA_DIR / PROMETHEUS_MEMORY_DB_PATH: default storage location.
• PROMETHEUS_MEMORY_NO_EMBEDDINGS=1: force FTS-only recall.
• PROMETHEUS_MEMORY_EMBEDDING_MODEL: defaults to BAAI/bge-small-en-v1.5.
• PROMETHEUS_MEMORY_EMBEDDING_API_URL and PROMETHEUS_MEMORY_EMBEDDING_API_KEY: OpenAI-compatible embedding endpoint.
• PROMETHEUS_MEMORY_LLM_ENABLED=1, PROMETHEUS_MEMORY_LLM_BASE_URL, PROMETHEUS_MEMORY_LLM_API_KEY, PROMETHEUS_MEMORY_LLM_MODEL: OpenAI-compatible LLM endpoint.

Local embeddings use the fastembed npm package. Its default BGESmallENV15 model is 384-dimensional and uses the package's CLS pooling plus vector normalization path. Local GGUF LLMs are not available in this package.

Commands

prometheus-memory remember "Use stable-cluster for production deploys"
prometheus-memory recall "production deploy target"
prometheus-memory stats
prometheus-memory sleep

Tests

bun --cwd packages/mnemopi test
bun --cwd packages/mnemopi run check