mengram-ai

JavaScript / TypeScript SDK for Mengram — Human-like memory for AI with 3 memory types: semantic, episodic, and procedural.

Install

npm install mengram-ai

Quick Start

const { MengramClient } = require('mengram-ai');

const m = new MengramClient('om-your-api-key');

// Add memories — auto-extracts facts, events, workflows
await m.add([
  { role: 'user', content: 'Fixed the auth bug. My process: check logs, reproduce locally, fix and deploy.' },
]);

// Semantic search (classic)
const results = await m.search('auth issues');

// Episodic — what happened?
const events = await m.episodes({ query: 'auth bug' });
// → [{summary: "Fixed auth bug", outcome: "Resolved", participants: [...]}]

// Procedural — how to do it?
const procs = await m.procedures({ query: 'debug' });
// → [{name: "Debug process", steps: [...], success_count: 3}]

// Unified search — all 3 types at once
const all = await m.searchAll('deployment issues');
// → { semantic: [...], episodic: [...], procedural: [...] }

// Procedure feedback — AI learns what works
await m.procedureFeedback(procId, { success: true });

// Experience-driven evolution — procedure improves on failure
await m.procedureFeedback(procId, {
  success: false, context: 'OOM on step 3', failedAtStep: 3
});

// View procedure version history
const history = await m.procedureHistory(procId);
// → { versions: [v1, v2, v3], evolution_log: [...] }

// Cognitive Profile — instant personalization
const profile = await m.getProfile('ali');
// → { system_prompt: "You are talking to Ali, a developer..." }

TypeScript

import { MengramClient, SearchResult, Episode, Procedure, UnifiedSearchResult, ProcedureHistoryResult } from 'mengram-ai';

const m = new MengramClient('om-...');

const results: SearchResult[] = await m.search('preferences');
const events: Episode[] = await m.episodes({ query: 'deployment' });
const procs: Procedure[] = await m.procedures({ query: 'release' });
const all: UnifiedSearchResult = await m.searchAll('issues');

Import Existing Data

Kill the cold-start problem — import ChatGPT, Obsidian, or text files (Node.js only, requires jszip for ChatGPT):

// ChatGPT export
npm install jszip  // one-time dependency
await m.importChatgpt('~/Downloads/chatgpt-export.zip');

// Obsidian vault
await m.importObsidian('~/Documents/MyVault');

// Text/markdown files
await m.importFiles(['notes.md', 'journal.txt']);

// With progress callback
await m.importChatgpt('export.zip', {
  onProgress: (current, total, title) => console.log(`${current}/${total} ${title}`)
});

API

| Method | Description | |--------|-------------| | Core | | | add(messages, options?) | Add memories (extracts all 3 types) | | addText(text, options?) | Add memories from plain text | | search(query, options?) | Semantic search | | searchAll(query, options?) | Unified search (all 3 types) | | getAll(options?) | List all memories | | getAllFull(options?) | List all memories with full details | | get(name) | Get specific entity | | delete(name) | Delete entity | | stats(options?) | Usage statistics | | Memory Types | | | episodes(options?) | Search/list episodic memories | | procedures(options?) | Search/list procedural memories | | procedureFeedback(id, options?) | Record success/failure (triggers evolution) | | procedureHistory(id) | Version history + evolution log | | procedureEvolution(id) | Evolution log (what changed and why) | | getProfile(userId?, options?) | Cognitive Profile | | Memory Management | | | dedup(options?) | Find and merge duplicate entities | | dedupAll(options?) | Deduplicate facts across all entities | | dedupEntity(name, options?) | Deduplicate facts on specific entity | | merge(source, target, options?) | Merge two entities | | archiveFact(entity, fact, options?) | Archive a specific fact | | reindex(options?) | Re-embed all entities | | fixEntityType(name, type, options?) | Fix entity type classification | | Search & Discovery | | | graph(options?) | Get knowledge graph (nodes + edges) | | timeline(options?) | Search facts by time range | | feed(options?) | Activity feed | | Agents & Insights | | | runAgents(options?) | Run memory agents (curator, connector, digest) | | agentHistory(options?) | Agent run history | | agentStatus(options?) | Check which agents are due | | insights(options?) | AI reflections | | reflect(options?) | Trigger memory reflection | | Triggers | | | getTriggers(options?) | List smart triggers | | detectTriggers(userId, options?) | Detect triggers for a user | | processTriggers() | Fire all pending triggers | | dismissTrigger(id) | Dismiss a trigger | | Webhooks | | | createWebhook(url, options?) | Create a webhook | | listWebhooks() | List all webhooks | | updateWebhook(id, options?) | Update a webhook | | deleteWebhook(id) | Delete a webhook | | Teams | | | createTeam(name) | Create shared team | | joinTeam(code) | Join team | | listTeams() | List your teams | | shareMemory(entity, teamId) | Share with team | | unshareMemory(entity, teamId) | Make memory personal again | | teamMembers(teamId) | List team members | | leaveTeam(teamId) | Leave a team | | deleteTeam(teamId) | Delete a team (owner only) | | Import | | | importChatgpt(zipPath, options?) | Import ChatGPT export ZIP | | importObsidian(vaultPath, options?) | Import Obsidian vault | | importFiles(paths, options?) | Import text/markdown files | | Billing | | | getBilling() | Current plan and usage | | createCheckout(plan) | Create checkout session | | createPortal() | Manage subscription |

Multi-User Isolation

Building a multi-tenant app? Pass userId to isolate memories per end-user. One API key, many users — each sees only their own data:

// Each userId gets its own isolated memory space
await m.add([{ role: 'user', content: 'I prefer dark mode' }], { userId: 'alice' });
await m.add([{ role: 'user', content: 'I prefer light mode' }], { userId: 'bob' });

const alice = await m.searchAll('preferences', { userId: 'alice' });
// → Only Alice's memories (dark mode)

const bob = await m.searchAll('preferences', { userId: 'bob' });
// → Only Bob's memories (light mode)

const profile = await m.getProfile('alice');
// → Alice's cognitive profile

No userId? Everything works as before — defaults to a single shared memory space.

Links

• Website
• Documentation
• Python SDK
• OpenClaw Plugin
• GitHub