MemGPT
Revolutionize AI interactions with personalized, long-term memory capabilities.. [Contact for Pricing]
Last verified:
What is MemGPT?
MemGPT (now Letta) is a memory-first AI agent platform that enables large language models to manage their own memory, overcoming the fundamental limitation of fixed context windows. Born from UC Berkeley's Sky Computing Lab, the system uses virtual context management inspired by operating systems' hierarchical memory to create agents with unbounded context. These persistent agents can be taught through language and improve from experience, maintaining continuity across multiple sessions and devices.
Key features include persistent memory systems that store core memory, recall memory, and archival memory; background memory agents (dream agents) that transform prompts and skills over time; cross-model memory portability allowing users to transfer agent memories between different AI providers; and the ability to create deeply personalized agents with unique identities and expertise. The platform offers Letta Code as its primary product—a memory-first coding agent accessible via terminal, desktop app, or remote environments—with support for skills, subagents, MCP tools, and git-based context repositories.
MemGPT/Letta is designed for developers, AI researchers, and tech enthusiasts building personalized AI applications, perpetual chatbots, automation tools, and long-running tasks requiring persistent state. The platform supports model-agnostic backends including OpenAI, Anthropic, Google Gemini, and local models through Ollama, making it suitable for customer service agents, research assistants, coding agents, document analysis systems, and multi-agent teams.
MemGPT pricing
Pricing model: Freemium
Free tier: Limited number of total agents and LLM requests with rotating free models. Personal Plans (for individual hands-on use via Letta Code or chat): Pro at $20/month includes usage quota for open-weights models and Letta Auto, pay-as-you-go for additional models/overage, up to 20 stateful agents. Max Lite at $100/month includes usage quota across all frontier model providers, 5X higher Letta Auto limits, up to 50 stateful agents. Max at $200/month includes increased frontier model quota, 20X higher Letta Auto limits, early access to new features. API Plan (for teams building applications): $20/month base plus $0.10 per active agent/month, $0.00015/sec tool execution, unlimited agents, API key authentication, pay-as-you-go LLM usage. Enterprise: Volume-based pricing with increased quotas, role-based access control, SAML/OIDC SSO, dedicated support. All plans support bringing your own API keys.
MemGPT pros
- Solves LLM context window limitation with virtual context management
- Persistent memory across sessions enables perpetual chatbots
- Cross-model portability transfers memories between AI providers
- Background dream agents continuously learn during idle time
- Memory palace visualization shows agent's complete memory state
- Supports 20 stateful agents on Pro plan, 50 on Max Lite, unlimited on API
- Bring your own API keys (BYOK) supported on all plans
- Open-source code available for self-hosting and customization
- Model-agnostic supporting OpenAI, Anthropic, Gemini, Ollama, llama.cpp
- Git-based context repositories with versioning for coding agents
- Skill learning dynamically learns skills through experience
- Remote MCP tools and client-side tools for local computer actions
- Sleep-time compute enables reasoning during idle periods
- Multi-device access with teleportation across machines
- Hierarchical memory system mimics human cognitive architecture
MemGPT cons
- Free tier has limited agents and LLM requests with rotating free models
- Token budget constraints still restrict simultaneously active information
- Server-side tool execution costs $0.00015/sec on API Plan
- Paid plans relatively expensive at $20/$100/$200 per month tiers
- Complexity may overwhelm non-technical users
- Requires OAuth authentication for Personal Plan quotas
- Web search/fetch tools not free unlike other built-in tools
- Legacy Pro Plan users automatically migrated to API Plan causing confusion
Frequently asked questions about MemGPT
Can I use my own API keys with Letta?
Yes. All plans support bringing your own API keys (BYOK). When you connect your own keys via /connect in Letta Code, usage goes directly through your provider account instead of consuming Letta credits.
What is the difference between Personal Plans and the API Plan?
Personal Plans (Pro, Max Lite, Max) are for individual, hands-on use via Letta Code or the chat interface with monthly usage quotas that reset. They require OAuth authentication. The API Plan is for developers and teams building applications on top of the Letta API with automated workloads, using API key authentication with purely usage-based credit pricing and unlimited agents.
What happens when I reach my plan limit?
On Personal Plans, you can continue using Letta with pay-as-you-go pricing for additional models or overage. On the API Plan, all usage is pay-as-you-go. You'll be notified by email when approaching your limit.
What are credits in Letta?
Credits are a standard cost unit for resources in Letta, such as LLM inference and CPU cycles. Model requests consume credits at a rate depending on the model API pricing. Current model pricing can be viewed at app.letta.com/models.
How is tool execution charged?
Server-side tools on the Letta API incur a credit cost for CPU time at $0.00015/sec on the API Plan. Remote MCP tools are executed by the MCP provider with no credit cost. Letta built-in tools are free except for web search/fetch tools. Client-side tools like bash tools in Letta Code run on your machine with no credit cost.
What is Letta Auto?
letta/auto and letta/auto-fast are model handles that automatically route to models optimized for Letta Code, recommended for best cost and performance. Personal Plans include usage quotas specifically for Letta Auto, with Max Lite having 5X higher limits than Pro and Max having 20X higher limits.
What are the usage limits for my plan?
The free tier includes a limited number of total agents and LLM requests with rotating free models. Personal Plans include monthly usage quotas that scale with each tier. Max Lite includes 5X the Letta Auto limits of Pro, and Max includes 20X. Paid plan users can view current usage in the account dashboard.
How does MemGPT's memory architecture work?
MemGPT uses virtual context management inspired by operating systems, creating a hierarchy with main context (like RAM) and external context (like disk). Core Memory contains always-accessible compressed essential facts, Recall Memory is a searchable database for reconstructing specific memories through semantic search, and Archival Memory provides long-term storage. The LLM itself acts as memory manager through self-directed memory editing via tool calling.
Can I create agents with distinct personas?
Yes. MemGPT can create and manage AI agents with distinct personas, transforming it from a memory management system into a robust agentic framework. Developers can create agents with specific roles, knowledge bases, and behavioral traits, each initialized with a unique persona guiding interactions and decision-making over time.
Where can I ask more questions or get support?
You can reach out to [email protected] for questions, or join the community on Discord for discussion and support with other users and developers.