2 lines of code · 9 connectors · 1,800+ tests

Give any AI agent
persistent, intelligent memory

Store and recall — that’s it. Auto-classification, entity extraction, knowledge graph, smart dedup, and hybrid retrieval are all built in. SDKs for Python, TypeScript, Go, and MCP.

$ pip install mindglue-sdk
$ npm install mindglue
$ go get github.com/mindglue/sdk-go
from mindglue_sdk import MindGlue

mg = MindGlue(api_key="mg_...")
mg.store("user:123", "Prefers dark mode, uses Python")

result = mg.recall("user:123", query="user preferences")
# Returns formatted markdown context ready for system prompt
8
Pipeline Steps per Store
3
Retrieval Signals Fused
<200ms
Avg Recall Latency
9
Data Source Connectors

How It Works

Three steps. No infrastructure to manage. Your agent gets smarter with every interaction.

1

Store

Your agent stores memories with a single API call. MindGlue auto-classifies the type, extracts entities and relationships, generates embeddings, and deduplicates — all automatically.

2

Recall

When your agent needs context, hybrid retrieval combines vector similarity, knowledge graph traversal, and temporal recency to find exactly the right memories.

3

Use

Get pre-formatted markdown context organized by type — facts, preferences, history, procedures — ready to inject directly into your system prompt.

One Call. Eight Intelligent Steps.

When you call store(), MindGlue runs a full intelligence pipeline automatically. This is what you’d spend months building yourself.

1

Classify

Auto-categorizes as fact, preference, episode, or procedure via LLM

2

Extract Entities

Identifies people, companies, products, and their relationships

3

Build Knowledge Graph

Entities and relationships become graph nodes with cross-source bridges

4

Generate Embeddings

Vector embeddings for semantic similarity search via HNSW index

5

Deduplicate

Catches paraphrases across sources, merges with contradiction detection

6

Normalize Entities

“Acme Corp” and “Acme Corporation” become the same canonical entity

7

Version & Audit

Full version history on every update, immutable audit trail

8

Persist & Cache

Stored in Postgres with pgvector, cached in Redis, webhook fired

Steps 1–4 run in parallel. The entire pipeline completes in under a second.

Why MindGlue?

Building memory from scratch means wiring together embeddings, a vector DB, a graph DB, caching, dedup logic, and ranking algorithms. Or you could use MindGlue.

Auto-Classification

Every memory is classified as a fact, preference, episode, or procedure — automatically via LLM with confidence scoring. No manual labeling. No taxonomy to maintain.

Hybrid Retrieval

Not just vector search. Three signals — semantic similarity, knowledge graph traversal, and temporal recency — fused into one ranked result. No other memory API does this.

Multi-Entity Recall

Recall memories across users, projects, and deals simultaneously. Perfect for complex agent interactions.

Intelligent Dedup

Paraphrases caught across sources. Duplicates merged via LLM. Contradictions detected automatically. Full version history preserved. Your memory stays clean without any effort.

Namespace Isolation

Full multi-tenant isolation. Each namespace is a separate memory universe with its own quotas and access control.

SDKs + MCP

Python, TypeScript, and Go SDKs. MCP server for Claude Desktop & Cursor. Real-time WebSocket streaming. Full REST API.

Connect Your Data Sources

Automatically sync data from the tools your team already uses. Every record flows through the full intelligence pipeline.

PostgreSQL

PostgreSQL

Database rows

Google Drive

Google Drive

Docs, Sheets, PDFs

Notion

Pages & databases

Slack

Channel messages

Salesforce

CRM records

Zendesk

Tickets & articles

GitHub

Issues & PRs

Jira

Jira

Project issues

Confluence

Wiki pages

Cross-Source Recall — The Real Magic

Ask about “Greenleaf Health” and get the Salesforce deal, the Notion onboarding doc, Jira tickets, Slack conversations, and Confluence specs — all in one recall. MindGlue automatically builds knowledge graph bridges between entities across every source.

Scale plan includes 3 connectors. Additional connectors $19/mo each. Enterprise: unlimited.

Simple, Transparent Pricing

Start free. Scale as you grow. No hidden fees.

Free

$0/mo

For prototyping & personal projects

  • 10,000 memories
  • 1,000 recall calls/mo
  • 2 namespaces
  • 3 recall strategies
  • Auto-classification
  • Dedup & versioning
Start Free
Most Popular

Pro

$29/mo

For teams building production agents

  • 100,000 memories
  • 25,000 recall calls/mo
  • 10 namespaces
  • Tags & search
  • Analytics & webhooks
  • Export/Import
Get Pro

Scale

$199/mo

For production workloads at scale

  • Unlimited memories
  • 100,000 recall calls/mo
  • Unlimited namespaces
  • 3 data connectors
  • Field encryption
  • WebSocket streaming
Get Scale

Enterprise

Custom

For organizations with compliance needs

  • Everything in Scale
  • Unlimited connectors
  • SSO / SAML
  • SLA guarantee
  • Private cloud / VPC deployment
  • Dedicated support & onboarding
Talk to Sales

Overages: Pro $0.50/10K memories + $1.00/5K recall calls. Scale $0.75/5K recall calls. Free plan has hard limits.

All plans include: auto-classification, intelligent deduplication, 3 recall strategies, knowledge graph, GDPR-compliant forget, version history

Frequently Asked Questions

How is MindGlue different from a vector database?
Vector databases only do similarity search. MindGlue is a complete memory intelligence layer that does much more: it auto-classifies memories into types (facts, preferences, episodes, procedures), extracts entities and relationships into a knowledge graph, handles deduplication with contradiction detection, normalizes entities across sources, applies temporal decay, and returns pre-formatted context ready for your system prompt. You get hybrid retrieval combining vector similarity, graph traversal, and temporal recency — not just nearest-neighbor search.
What data sources can I connect?
MindGlue ships with 9 built-in connectors: Salesforce, Notion, Google Drive, Slack, Zendesk, PostgreSQL, GitHub, Jira, and Confluence. Each connector uses OAuth for secure one-click authentication, lets you choose exactly which resources to sync (repos, projects, channels, tables), and runs incremental syncs on your schedule. Data from every source flows through the full intelligence pipeline — classified, entity-extracted, deduplicated, and graph-linked — so your agent can recall cross-source context in a single query.
How does cross-source recall work?
When you store memories from multiple sources, MindGlue automatically builds bridge nodes in the knowledge graph linking entities across sources. Ask about "Greenleaf Health" and you'll get the Salesforce deal, the Notion onboarding doc, Jira tickets, Slack conversations, and Confluence specs — all in one recall. The recall engine uses three strategies (precise, balanced, broad), source-authority ranking so CRM data outweighs chat messages, and entity normalization so "Acme Corp" and "Acme Corporation" are understood as the same entity.
Does MindGlue work with ChatGPT, Claude, and local models?
Yes. MindGlue is model-agnostic — it works with any AI model or framework. Use our REST API, Python/TypeScript/Go SDKs, or the MCP server for Claude Desktop. The recall endpoint returns pre-formatted markdown context that you inject into any model's system prompt. Internally, MindGlue supports OpenAI, Anthropic Claude, and Ollama (fully local) as intelligence providers for classification, extraction, and embeddings.
What kinds of AI agents benefit from MindGlue?
Any agent that needs to remember things across conversations. Customer support bots that recall a user's full history. Sales copilots that know every deal, contact, and conversation. Engineering assistants that remember architecture decisions, sprint context, and incident postmortems. Product agents that connect user research to feature specs. If your agent would be better with context from past interactions or connected data sources, MindGlue is for you.
How does pricing work if I exceed my plan limits?
Free plan has hard limits — you'll get a clear error with an upgrade prompt. Paid plans (Pro and Scale) use soft limits: your agent keeps working and overages are billed at the rates shown on the pricing table. No surprise outages, no data loss. You can monitor your usage in real time from the dashboard.
Is my data secure?
Yes. Full namespace isolation ensures no cross-tenant data leaks. Scale plan includes optional field-level encryption (AES-256 via Fernet with per-namespace key derivation). GDPR-compliant forget endpoint permanently deletes data on request. API keys are SHA-256 hashed. Connector credentials are encrypted at rest. All connections use TLS. OAuth connectors use industry-standard HMAC-signed CSRF state and never store raw passwords. See our security page for full details.
Can MindGlue run in our own infrastructure?
Yes — our Enterprise plan includes private cloud and VPC deployment options for organizations with data residency, compliance, or air-gapped requirements. We handle the deployment, configuration, and ongoing maintenance in your environment so you get the full MindGlue platform without managing infrastructure. Talk to our sales team to discuss your deployment needs. For most teams, our managed cloud at app.mindglue.ai is the fastest way to get started — the Free tier includes 10K memories and 1K recall calls per month.

Ready to give your AI agent memory?

Start free. No credit card required. Upgrade when you need more.