Memori
Memori is agent-native memory infrastructure. A LLM-agnostic layer that turns agent execution and conversation into structured, persistent state for production systems.
Last verified:
What is Memori?
Memori is an open-source, agent-native memory infrastructure that gives AI agents a structured, persistent memory layer. It automatically captures conversations, extracts meaningful facts, preferences, rules, and summaries, and makes them searchable across entities, processes, and sessions. Memori uses Advanced Augmentation to automatically enhance memories with facts, preferences, attributes, events, and relationships while running in the background with zero added latency.
Key features include targeted recall that pulls only relevant context across conversations, selective semantic search that enriches searches with semantic context without ballooning token costs, explainable results with clear lineage showing why something was included, a knowledge graph showing how memories connect, analytics tracking memory creation and recall usage, and a payments vault that is PCI and SOC 2 compliant for safely using cards and PII. Memori scores 81.95% accuracy on LoCoMo benchmarks while reducing token usage by 95% compared to full-context retrieval.
Memori is designed for developers building AI applications, agents, and copilots that need long-term memory. It works with enterprise teams needing production-ready memory infrastructure, customer support AI systems, personal assistants, robotics applications like NAMUHx wellness robots, and any LLM application requiring persistent memory. It supports OpenAI, Anthropic, Bedrock, Gemini, Grok, LangChain, Pydantic AI, and works with PostgreSQL, MySQL, SQLite, MongoDB, CockroachDB, Neon, Supabase, Oracle, and more.
Memori offers two deployment options: Memori Cloud (fully hosted with managed storage and built-in observability) and Memori BYODB (bring your own database for full infrastructure control). The SDK is the same for both, allowing teams to start with Cloud for speed and migrate to BYODB if needed.
Memori pricing
Pricing model: Freemium
Memori offers a free open-source version you can self-host with your own database. Memori Cloud launches with a free tier including free starting credits for testing and intermittent use. Teams can sign up at app.memorilabs.ai for a free API key. There is a Pro plan available - startups can apply for Memori's startup program to get free access to the 3-month Pro plan. CockroachDB integration mentions free starting credits perfect for testing without upfront database costs. Pricing details for paid tiers beyond the free tier and startup program are not explicitly published on the website - teams needing production credits or enterprise features would need to contact sales or sign up to see plan details.
Memori pros
- Open source system with transparent code
- SQL-native memory layer using existing infrastructure
- Automatically captures conversations without code changes
- Advanced Augmentation runs in background with zero latency
- Database, LLM, and framework agnostic integration
- Reduces token usage by 95% compared to full-context retrieval
- 81.95% accuracy on LoCoMo benchmark testing
- Plugs directly into existing systems without rewriting code
- Supports OpenAI, Anthropic, Bedrock, Gemini, Grok all modes
- Native LangChain and Pydantic AI framework support
- Knowledge graph view for relationship-first exploration
- Explainable results with clear lineage and relevance tracing
- PCI and SOC 2 compliant payments vault for PII safety
- Role-based access control with SSO and OAuth integration
- Automatic data retention policies with audit trail
- Memori Cloud enables production memory in minutes with API key
- Cut LLM costs by over 95% with tokenless recall
- Milliseconds response times vs seconds with caching
- Dual memory system with Conscious and Auto ingest modes
- Multi-tenant support with namespace separation
Memori cons
- Memories do not automatically transfer between Cloud and BYODB environments
- Requires Python 3.10 or higher for quick start
- Need separate API key from OpenAI or other LLM provider
- BYODB requires self-managed operations and infrastructure
- Advanced Augmentation runs asynchronously which may delay memory creation
- For CLI programs need to manually wait for augmentation to finish
- Cloud environment quota limits may require plan upgrade
- Knowledge graph exploration has learning curve for new users
Frequently asked questions about Memori
What is Memori and what does it do?
Memori is an open source system that gives your AI agents a structured, persistent memory layer. It automatically captures conversations, extracts meaningful facts, and makes them searchable across entities, processes, and sessions. Memori uses Advanced Augmentation to automatically enhance memories with facts, preferences, attributes, events, relationships, and more while running in the background with no added latency.
How do I get started with Memori?
You can get started in under 3 minutes. Install Memori with pip install memori, set your OpenAI API key as an environment variable, create a Python file with the Memori SDK, and run your first application. For Memori Cloud, sign up at app.memorilabs.ai for a free API key and initialize Memori() - no database provisioning needed. The quick start uses OpenAI, SQLAlchemy, and SQLite as an example.
What is the difference between Memori Cloud and BYODB?
Memori Cloud is fully hosted with managed storage and built-in observability - best for teams wanting the fastest path to production with minimal ops overhead. You just set MEMORI_API_KEY and initialize Memori(). BYODB (Bring Your Own Database) gives full control over storage and infrastructure using your own database - best for teams needing their own deployment boundary and self-managed operations. The SDK is the same for both, but memories do not automatically transfer between environments.
What databases does Memori support?
Memori supports PostgreSQL, MySQL/MariaDB, OceanBase, SQLite, MongoDB, CockroachDB, Neon, Supabase, Oracle, and more through DB API 2.0, SQLAlchemy, and Django ORM. It is SQL-native and works with cloud databases like Neon and Supabase as well as serverless options. You can use your existing database infrastructure you already know how to scale.
Which LLM providers and frameworks work with Memori?
Memori supports OpenAI, Anthropic, Bedrock, Gemini, and Grok (xAI) in all modes including streamed, unstreamed, sync, and async. It has native support for LangChain and Pydantic AI frameworks and works with Nebius AI Studio. Through LiteLLM integration it works with ANY LLM library including Azure OpenAI, custom providers, and Ollama without monkey-patching overhead.
How does Memori reduce costs and improve performance?
Memori cuts costs by over 95% through tokenless recall, structured memory from trace that prevents agents from repeating mistakes, and intelligent routing. It reduces token usage by 95% compared to full-context retrieval (tested on LoCoMo benchmark). Caching concise snippets along the recall path reduces round-trips and recomputation, resulting in consistently millisecond responses vs seconds. Traditional context injection uses 2000+ tokens while Memori uses ~150-300 tokens.
What is Advanced Augmentation?
Advanced Augmentation is Memori's intelligent background system that automatically enhances memories with facts, preferences, attributes, events, relationships, and more. It runs asynchronously so there is no added latency to AI processes. It extracts structured memory from agent traces including execution paths, tool results, workflow steps, outcomes, and decision-making logic rather than relying primarily on long-form conversation history.
How do I inspect and debug memories?
Use the Memories page in app.memorilabs.ai to inspect all memories created by your applications. You can view data in Table view for row-based inspection or Knowledge graph view for relationship-first exploration. The Memories tab shows summary, entity ID, recalls, mentions, last retrieved, and created at. Click any row to open Memory details panel with memory ID, importance score, and associated subjects. Switch to Knowledge graph to explore entities and trace cross-entity relationships.
What memory types does Memori categorize?
Memori automatically categorizes memory into five types: Facts (objective information like I use PostgreSQL), Preferences (personal choices like preferring clean code), Skills (abilities and expertise like FastAPI experience), Context (project information like working on e-commerce platform), and Rules (guidelines and constraints like always write tests first). You can search memories by category and set different retention policies for short-term, long-term, and permanent memory.
How secure is Memori and what data controls exist?
Memori offers role-based access control (RBAC) with read, write, and admin access roles, integration with identity providers (SSO, OAuth), data retention and deletion policies with automatic purging and full audit trail, audit and observability tracking who accessed what when and why, and data ownership where your data stays in your database. Memori never moves, stores, or indexes data outside your chosen environment unless you explicitly enable Memori Cloud. The Payments Vault is PCI and SOC 2 compliant for secure memory enabling agents to safely use cards and PII.