mengram-ai 2.5.4

Human-like memory for AI — semantic, episodic & procedural. Cognitive Profile, unified search, memory agents. Free open-source Mem0 alternative.

Mengram — Human-Like Memory for AI

The only AI memory API with 3 memory types: semantic, episodic, and procedural. Your AI remembers facts, events, and learned workflows — just like a human brain.

Website · Dashboard · API Docs · PyPI · npm

---

Why Mengram?

	Mengram	Mem0	Supermemory
Semantic Memory (facts)	✅	✅	✅
Episodic Memory (events)	✅	❌	❌
Procedural Memory (workflows)	✅	❌	❌
Cognitive Profile	✅	❌	❌
Unified Search (all 3 types)	✅	❌	❌
Knowledge Graph	✅	✅	❌
Autonomous Agents	✅ Curator, Connector, Digest	❌	❌
Team Shared Memory	✅	❌	✅
AI Reflections	✅	❌	❌
Webhooks	✅	✅	✅
MCP Server	✅ Claude Desktop, Cursor, Windsurf	✅	❌
LangChain Integration	✅	❌	❌
Python & JS SDK	✅	✅	✅
Self-hostable	✅	✅	✅
Price	Free	$19-249/mo	Enterprise

3 Memory Types

Mengram automatically extracts all 3 types from a single add() call:

🧠 Semantic — Facts, preferences, skills: "uses Python", "prefers dark mode"

📝 Episodic — Events, decisions, experiences: "Debugged Railway deployment for 3 hours, fixed pgvector issue"

⚙️ Procedural — Learned workflows, processes: "Deploy: build → twine upload → npm publish → git push"

# One call extracts all 3 types automatically
m.add([
    {"role": "user", "content": "Fixed the auth bug today. Problem was API key cache TTL. My debug process: check Railway logs, reproduce locally, fix and deploy."},
])
# → Semantic: "API key caching caused auth bug"
# → Episodic: "Debugged auth bug, fixed cache TTL"  
# → Procedural: "Debug process: logs → reproduce → fix → deploy"

Quick Start (60 seconds)

1. Get API key

Sign up at mengram.io — free, no credit card.

2. Install

pip install mengram-ai    # Python
npm install mengram-ai    # JavaScript / TypeScript

3. Connect to Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "mengram": {
      "command": "mengram",
      "args": ["server", "--cloud"],
      "env": {
        "MENGRAM_API_KEY": "your-key-here"
      }
    }
  }
}

Done. Claude now has persistent memory with all 3 types.

Python SDK

from mengram.cloud.client import CloudMemory

m = CloudMemory(api_key="om-...")

# Add memories — auto-extracts facts, events, workflows
m.add([
    {"role": "user", "content": "I deployed Mengram on Railway with PostgreSQL 15"},
    {"role": "assistant", "content": "Great, noted the deployment setup."}
], user_id="ali")

# Semantic search (classic)
results = m.search("deployment setup", user_id="ali")

# Episodic search — what happened?
events = m.episodes(query="deployment", user_id="ali")
# → [{summary: "Deployed on Railway", outcome: "Success", participants: [...]}]

# Procedural search — how to do it?
procs = m.procedures(query="deploy", user_id="ali")
# → [{name: "Deploy Mengram", steps: [...], success_count: 5}]

# Unified search — all 3 types at once
all_results = m.search_all("deployment issues", user_id="ali")
# → {semantic: [...], episodic: [...], procedural: [...]}

# Procedure feedback — AI learns what works
m.procedure_feedback(proc_id, success=True)

# Cognitive Profile — instant personalization
profile = m.get_profile("ali")
# → {system_prompt: "You are talking to Ali, a developer in Almaty..."}

# Use profile with any LLM
import openai
response = openai.chat.completions.create(
    model="gpt-4",
    messages=[
        {"role": "system", "content": profile["system_prompt"]},
        {"role": "user", "content": "What should I work on next?"}
    ]
)

# Memory agents
m.run_agents(agent="all", auto_fix=True)

# Team memory
team = m.create_team("Backend Team")
m.share_memory("Redis", team_id=team["id"])

JavaScript / TypeScript SDK

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

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

// Add memories — extracts all 3 types
await m.add([
  { role: 'user', content: 'Fixed OOM with Redis cache' },
], { userId: 'ali' });

// Episodic — what happened?
const events = await m.episodes({ query: 'OOM fix' });

// Procedural — how to do it?
const procs = await m.procedures({ query: 'cache setup' });

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

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

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

Full TypeScript types included with Episode, Procedure, and UnifiedSearchResult interfaces.

Cognitive Profile

One API call generates a ready-to-use system prompt from all 3 memory types:

profile = m.get_profile("ali")
print(profile["system_prompt"])

Output:

You are talking to Ali, a 22-year-old developer in Almaty building Mengram.
He uses Python, PostgreSQL, and Railway. Recently: debugged pgvector deployment,
researched competitors Mem0 and Supermemory, designed freemium pricing.
Workflows: deploys via build→twine→npm→git, prefers iterative shipping.
Communicate in Russian/English, direct style, focus on practical next steps.

Insert into any LLM's system prompt for instant personalization. Replace your RAG pipeline.

LangChain Integration

Drop-in replacement for LangChain's memory. Instead of returning raw message history, Mengram returns relevant knowledge from all 3 memory types.

pip install mengram-ai[langchain]

LCEL (recommended):

from mengram.integrations.langchain import MengramChatMessageHistory
from langchain_core.runnables.history import RunnableWithMessageHistory

chain_with_memory = RunnableWithMessageHistory(
    chain,
    lambda session_id: MengramChatMessageHistory(
        api_key="om-...", session_id=session_id, user_id="ali"
    ),
    input_messages_key="input",
    history_messages_key="history",
)

ConversationChain (legacy):

from mengram.integrations.langchain import MengramMemory

# Basic — search-based context
memory = MengramMemory(api_key="om-...", user_id="ali")

# With Cognitive Profile — full user personalization
memory = MengramMemory(api_key="om-...", user_id="ali", use_profile=True)

chain = ConversationChain(llm=llm, memory=memory)
chain.predict(input="I deployed my app on Railway")
# Next call — Mengram searches all 3 memory types for relevant context
chain.predict(input="How did my last deployment go?")
# → Memory provides: facts about Railway, the deployment event, deploy workflow

vs ConversationBufferMemory:

	ConversationBufferMemory	MengramMemory
Storage	RAM (lost on restart)	Persistent (PostgreSQL)
Context	Last N messages (raw)	Relevant knowledge (semantic search)
Memory types	1 (messages)	3 (semantic + episodic + procedural)
Cross-session	❌	✅
Personalization	❌	✅ Cognitive Profile

Memory Categories

Separate memory by user, agent, session, and application:

m.add(messages, user_id="ali")                           # User's memory
m.add(messages, user_id="ali", agent_id="support-bot")   # Agent's memory
m.add(messages, user_id="ali", run_id="session-123")     # Session-scoped
m.add(messages, user_id="ali", app_id="helpdesk")        # App-scoped

Memory Agents

Three autonomous agents that analyze your memory:

🧹 Curator — Finds contradictions, stale facts, duplicates. Auto-cleans with auto_fix=True.

🔗 Connector — Discovers hidden connections, behavioral patterns, skill clusters.

📰 Digest — Weekly summary with headlines, trends, and recommendations.

API Endpoints

Endpoint	Description
POST /v1/add	Add memories (auto-extracts all 3 types)
POST /v1/search	Semantic search
POST /v1/search/all	Unified search (semantic + episodic + procedural)
GET /v1/episodes	List episodic memories
GET /v1/episodes/search	Search episodes by meaning
GET /v1/procedures	List procedural memories
GET /v1/procedures/search	Search procedures by trigger
PATCH /v1/procedures/{id}/feedback	Record success/failure
GET /v1/profile	Cognitive Profile (system prompt)
GET /v1/profile/{user_id}	Profile for specific user
POST /v1/agents/run	Run memory agents
GET /v1/insights	AI-generated insights
GET /v1/graph	Knowledge graph
GET /v1/timeline	Temporal search
POST /v1/teams	Create team
POST /v1/webhooks	Create webhook
GET /v1/keys	List API keys
GET /v1/stats	Usage statistics

Full docs: https://mengram.io/docs

Architecture

┌──────────────────────────────────────┐
│          Your AI Clients             │
│  Claude Desktop · Cursor · Windsurf  │
└──────────────┬───────────────────────┘
               │ MCP / REST API
┌──────────────▼───────────────────────┐
│        Mengram Cloud API             │
│  Extraction · Re-ranking · Search    │
├──────────────────────────────────────┤
│       3 Memory Types                 │
│  🧠 Semantic · 📝 Episodic · ⚙️ Proc │
├──────────────────────────────────────┤
│       Memory Agents Layer            │
│  🧹 Curator · 🔗 Connector · 📰 Digest│
├──────────────────────────────────────┤
│       Storage Layer                  │
│  PostgreSQL · pgvector · Teams       │
│  Webhooks · Reflections · Graph      │
└──────────────────────────────────────┘

License

Apache 2.0

---

Built by Ali Baizhanov