Agent Memory / Context Management
Tool | Category | Segment | Platform / Tool | Plan / License | Monthly Price USD | Pricing Model | Free Tier / OSS | Included Usage / Limits | Memory Model / Types | Persistence / Retrieval | Context Management / Personalization | Integrations / Frameworks | Deployment / Hosting | Security / Privacy | Team / Governance | Best Fit | Main Limits / Caveats |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
No tagline | Agent Memory / Context Management | Stateful agent runtime | Letta API / Letta Code | Open source plus Letta Cloud | $20/month API plan; personal Pro $20/month; free BYOK/local path | Usage-based API plan plus personal quota plans | ✓ | Free Letta Code can run with BYOK/local models; free Constellation accounts support up to three agents with managed state; API plan adds $0.10 per active agent/month and $0.00015/sec tool execution | MemGPT-style self-editing memory blocks, messages, conversations, MemFS context repositories and sleep-time/reflection concepts | All agent state including memories, user messages, reasoning and tool calls is persisted in a database; MemFS stores git-backed markdown memory | Agents can edit their own memory and keep core memories in context while retrieving older state after compaction | Letta API, Letta Code, MCP tools, server/client tools, OpenAI/Anthropic/OpenRouter/BYOK providers | Local open-source, Docker server, Letta Cloud/Constellation and remote environments | Self-hosting supports local control; Cloud features require account; BYOK supported | Team/Enterprise plans available; Enterprise adds RBAC, SAML/OIDC SSO and dedicated support | Stateful agents that need inspectable memory, versioned context and agent self-management | API and personal plans are distinct; Docker server can use legacy memory blocks instead of MemFS |
No tagline | Agent Memory / Context Management | Temporal graph memory SaaS | Zep | Commercial cloud SaaS | $0 free; Flex $125/month; Flex Plus $375/month | Credit-based plans by episode size; retrieval, storage and users are unmetered | ✓ | Free plan includes 1,000 credits/month, 2 projects, 5 custom entity/edge types and variable rate limits; Flex includes 50,000 credits/month and 600 RPM | Temporal Context Graphs with episodes, entities, facts, relationships, custom entity/edge types and observations on higher plans | Ingests chat messages, JSON payloads and text into a context graph; retrieval, storage, threads and users cost 0 credits | Context Lake serves long-running agent memory with temporal facts and production context assembly | Python, TypeScript and Go SDKs; Graphiti engine; common agent framework integrations | Zep Cloud, Cloud plus BYOK and Enterprise BYOC deployment options | SOC 2 Type II on paid/cloud paths; Enterprise adds HIPAA BAA, audit logs and longer API log retention | Free/Flex project limits; Enterprise adds custom credits, SLA, unlimited projects and dedicated support | Production agents needing managed temporal memory and graph-based recall at scale | Ingestion credits scale with episode size; free processing priority and rate limits are variable |
No tagline | Agent Memory / Context Management | Cloud provider agent memory | Amazon Bedrock AgentCore Memory | AWS managed service | Usage-based | Short-term events, long-term storage and retrieval pricing | No durable free tier captured | Short-term memory is $0.25 per 1,000 new events; long-term storage is $0.75 per 1,000 records/month with built-in strategies or $0.25 with override/self-managed strategies; retrieval is $0.50 per 1,000 record retrievals | Short-term memory for multi-turn conversations and long-term memory that persists across sessions, with shared memory stores across agents | AgentCore Memory stores events/records and supports retrieval through AWS managed APIs | Provides context-aware agents with control over what agents remember and learn | LangGraph, LangChain, Strands and LlamaIndex | AWS managed AgentCore service | AWS IAM, regional service controls and AWS content/data policies apply | AWS account governance, IAM, CloudWatch observability and service quotas | AWS-centered teams building production agents with managed memory and enterprise controls | Charges combine memory plus model and other AgentCore components; availability and regions need confirmation |
No tagline | Agent Memory / Context Management | Learning memory system | Hindsight by Vectorize | MIT open source plus managed cloud | $0 self-hosted; Cloud pay-as-you-go | Self-host free or usage-based token billing | ✓ | Self-host includes all four memory networks, retain/recall/reflect APIs, MCP server and embedded PostgreSQL; Cloud starts with free credits | Four memory networks with Retain, Recall, Reflect, Iris Extract and Mental Model operations | Stores, retrieves and synthesizes memory from agent experience; Cloud provides managed infrastructure and backups | Aims at agents that learn from repeated experience instead of only recalling snippets | REST API, Python SDK, MCP, LangChain, CrewAI, LlamaIndex and other integrations | Self-hosted Docker or Hindsight Cloud | Self-host keeps data on customer infrastructure; Enterprise supports BYOC/on-prem and custom SLA | Cloud adds dashboard, usage analytics, team collaboration and support SLA; Enterprise adds SSO/RBAC | Agents that must retain lessons, failures and behavioral patterns across sessions | Cloud costs vary by operation tokens; self-host operations still require model/infrastructure spend |
No tagline | Agent Memory / Context Management | Coding-agent memory MCP | OpenMemory by Mem0 | Mem0 product / MCP memory server | $0 software path; Mem0/API costs may apply | Memory MCP layer for coding agents | ✓ | Public page presents install path and project-scoped memory; exact hosted quotas follow Mem0 account/API setup | Typed coding memories such as preferences and implementation details, tagged by project/repo | Auto-captures, organizes, searches and injects relevant memory into coding agents | Feeds project-specific context automatically so agents do not need repeated manual prompting | MCP-compatible coding agents, IDEs and Mem0 ecosystem | Local/plugin plus Mem0-backed services depending configuration | Access logs show memory added, edited or served; visibility rules and tags are documented on product page | Individual/project memory management; broader governance follows Mem0 account controls | Developers wanting portable memory across Claude Code, Cursor, Codex-like agents and IDE sessions | Not a general backend memory database by itself; exact data path and model/API costs depend on setup |
No tagline | Agent Memory / Context Management | Cloud provider sessions and memory | Vertex AI Agent Engine Sessions / Memory Bank | Google Cloud pre-GA feature | Usage-based; sessions/memory bank billing from 2026-01-28 | Event-based pricing plus Agent Engine runtime resources | No public free tier captured beyond Google account/Express Mode paths | Pricing page states billing begins for Code Execution, Sessions and Memory Bank on 2026-01-28; public search result reports Sessions and Memory Bank at $0.25 per 1,000 events | Sessions hold event history and state; Memory is personalized information accessed across multiple sessions for a user | Sessions maintain chronological events and can be managed via ADK or API calls | Supports cross-session continuity and personalization for deployed agents | Google ADK, Vertex AI Agent Engine SDK and direct API calls | Google Cloud Vertex AI Agent Engine | Pre-GA terms apply; Google Cloud IAM, project and region controls apply | Project-level IAM, quotas and billing governance | Google Cloud teams deploying ADK/Vertex agents that need managed session state and user memory | Pre-GA status means behavior, support and pricing can change; exact regional pricing should be checked in Cloud pricing pages |
No tagline | Agent Memory / Context Management | Managed memory layer | Mem0 Platform | Hosted SaaS plus open-source SDK | $0 Hobby; paid from $19/month | Plan-based requests plus custom usage-based pricing | ✓ | Hobby includes unlimited end users, 10,000 add requests/month, 1,000 retrieval requests/month and 1 project; Starter is $19/month with 50,000 add and 5,000 retrieval requests | Conversation, session, user and organizational memory layers; graph memory and vector-backed memory depending mode | Managed platform stores, enriches and retrieves memories with vector store, graph services and rerankers | Designed for persistent personalization so agents remember user and org context without prompt bloat | LangChain, CrewAI, Vercel AI SDK, REST API, Python/JS SDKs and 20+ integrations | Mem0 Cloud or self-hosted open-source stack | Docs warn to avoid storing secrets or unredacted PII; platform docs mention audit logs and workspace governance | Project limits by plan; Enterprise adds unlimited usage, SSO, custom integrations and audit logs | Teams needing a production memory API that can be added to agents with minimal code | Graph and advanced governance differ between open-source and platform modes; request limits can be tight on free tier |
No tagline | Agent Memory / Context Management | Provider conversation state | OpenAI Conversations API | OpenAI API feature | Model/API usage based | Conversation state object used with Responses API | Included as API capability | No standalone free memory tier captured; Responses/model usage and storage/data-retention policies apply | Conversation objects store messages, tool calls, tool outputs and other items under a durable identifier | Subsequent Responses calls can reference the conversation so the platform prepends stored items | Simplifies multi-turn state across sessions, devices or jobs without sending full history manually | OpenAI Responses API, OpenAI SDKs and custom apps | Hosted OpenAI API | OpenAI API data handling and organization settings apply; zero data retention changes stateful options | Organization/project/API key controls and application-managed deletion/retention patterns | Apps that need hosted conversation continuity but not a full semantic memory layer | Conversation history is not the same as curated long-term memory; apps still need compaction, retention and relevance policies |
No tagline | Agent Memory / Context Management | Agent framework memory service | Google ADK Memory | Open-source framework feature | $0 SDK; backend/model costs separate | Memory service interface with local and Vertex backends | ✓ | No ADK software cap; examples can use local memory or VertexAiMemoryBankService with Google Cloud billing | Session memory and long-term memory via Memory Bank; memories can be saved from sessions and loaded into agents | ADK server connects to a memory service URI; callbacks can auto-save sessions to memory | Gives ADK agents cross-session recall and personalization when paired with Memory Bank | Google ADK, VertexAiMemoryBankService, ADK web/api_server | Local ADK app or Vertex AI Agent Engine | Data path depends on selected memory service; Vertex path follows Google Cloud terms | Governance follows local deployment or Google Cloud project controls | Teams using Google ADK who want memory without adopting a separate memory vendor | Memory quality depends on configured service and callbacks; managed features are tied to Vertex AI maturity and billing |
No tagline | Agent Memory / Context Management | Agent memory and RAG | Microsoft AutoGen Memory | Open-source framework feature | $0 software; storage/model costs separate | Framework memory abstraction with vector memory examples | ✓ | No software cap; ChromaDB extra and embedding models are needed for vector memory examples | Memory and RAG stores, including ChromaDBVectorMemory for semantic retrieval | Memory stores content and updates agent context with relevant retrieved items | Adds persistent context or RAG memory to AutoGen agents during conversations | AutoGen AgentChat, autogen-ext memory components, ChromaDB and custom Memory implementations | Application code / self-hosted | Data path depends on chosen vector DB and embedding provider | No hosted governance in OSS; app owns retention and tenant isolation | AutoGen users adding retrieval-backed memory to agents | Reference implementation is vector/RAG-oriented; advanced memory policies must be built around it |
No tagline | Agent Memory / Context Management | Personal agent memory hub | Membase | Commercial SaaS | $0 Free; Pro $20/month | Plan-based personal memory hub | ✓ | Free includes limited memory searches, episodes, wiki documents and AI chats; Pro adds unlimited memory searches and larger limits | Self-evolving personal memory with episodes, wiki documents, AI chats and MCP integration | Automatically builds personal memory from daily agents and apps | Keeps context across Cursor, ChatGPT, Claude Desktop, Claude Code, Codex, VS Code and MCP-compatible apps | MCP-compatible agents, Gmail, Calendar, Slack, ChatGPT/Claude/Gemini importers and coding tools | Hosted SaaS | Pricing page states opt out of model training and account deletion controls; full security details need account review | Free/Pro personal plan controls; community/priority support by plan | Power users wanting one personal memory across many AI tools | Less developer-backend-oriented than Mem0/Zep; limited public details on enterprise controls |
No tagline | Agent Memory / Context Management | Agent framework memory | Agno Memory | Open-source framework feature | $0 software; database/model costs separate | Database-backed framework memory | ✓ | No software cap; memories are stored in the connected database such as SQLite, Postgres or MongoDB | Automatic memory and agentic memory; chat history, user preferences and session summaries in v1 docs | Stores memories in agent database tables/collections and retrieves them on future runs | update_memory_on_run automatically extracts memories; enable_agentic_memory gives the agent tools to manage memory | Agno agents, teams, SQLite, Postgres, MongoDB and major LLM providers | Application code / self-hosted | Data stays in configured database; model provider choice controls memory extraction data path | Database and application controls; no standalone governance layer in core memory feature | Agno users who need simple persistent user memory in agent apps | Automatic and agentic memory modes are mutually exclusive; memory quality depends on LLM extraction |
No tagline | Agent Memory / Context Management | Open-source memory framework | MemoryScope | Open-source / token ecosystem | Public fixed price not captured | Package plus hosted ecosystem | ✓ | Public site links documentation and GitHub; exact free quota or paid pricing was not captured | Long-term memory capabilities for LLM chatbots and agents; collective experience in a MemoryScope | Package API creates/manages agents and interacts with their memories | Aims to give agents flexible long-term memory and shared experience | Python package, GitHub project and MemoryScope ecosystem | Package/local plus hosted ecosystem signals | Security/privacy details were not publicly itemized in captured source | Governance details not captured | Researchers/devs exploring a dedicated long-term memory framework from local seeds | Official pricing, hosting and enterprise controls are unclear; verify project maturity before relying on it |
No tagline | Agent Memory / Context Management | AI-native memory library | Honcho | Open source plus managed service | Public fixed price not captured | Memory library with managed service | ✓ | Docs describe open-source memory library and managed service; public pricing was not found in official docs search | Long-term memory about entities such as users, agents, groups and ideas; social intelligence over stored history | Stores history and provides tools to traverse it and infer latent context | Builds state about changing entities over time for personalized and social agents | Any model/framework/architecture through Honcho APIs/library | Self-hosted/open source or Honcho managed service | Data/privacy details require review of Honcho deployment docs and terms | Managed service governance not publicly priced in captured docs | Agents requiring rich identity, social context and entity-centered memory | Pricing and quotas are not clearly public; evaluate maturity and hosting model before production |
No tagline | Agent Memory / Context Management | Open-source cognitive memory | Engram | Apache-2.0 open source | $0 self-hosted; planned Developer $29/month; planned Team $149/month | Self-host free; managed cloud in development | ✓ | Self-hosted plan lists unlimited agents and memories with all 4 cognitive memory types; Developer and Team cloud plans are marked coming soon | Semantic, episodic, procedural and working memory with confidence scoring, decay, contradiction detection and lifecycle management | REST API stores and recalls typed memories with hybrid vector plus graph retrieval | Hot memories can auto-inject into prompts while cold memories surface only when relevant | Python SDK, REST API, LangChain integration, multi-LLM support and Docker Compose | Self-hosted Go/Postgres/pgvector; cloud waitlist | Self-hosting owns data completely; managed cloud not yet generally available | Community support in self-hosted; planned Team includes workspaces, dashboard and SLA | Teams wanting inspectable open-source cognitive memory with decay and contradiction handling | Cloud plans are not live; product/domain variants named Engram may be confusing and need validation |
No tagline | Agent Memory / Context Management | Edge platform agent memory | Cloudflare Agents Memory | Cloudflare platform feature | Workers/Durable Objects pricing applies | Platform memory feature for Cloudflare Agents | Included with platform capability | No separate memory price captured; Cloudflare Workers, storage and related platform billing apply | Conversation history plus context memory blocks that can be read, written, searched and loaded | Cloudflare Agents manage persisted session state and context memory blocks | Keeps agent conversations and reusable context available inside Cloudflare edge-hosted agents | Cloudflare Agents SDK, Workers, Durable Objects and platform storage | Cloudflare edge platform | Cloudflare account/security controls and platform data policies apply | Cloudflare project/team controls and deployment governance | Developers building agent apps directly on Cloudflare's platform | Platform-specific; not a vendor-neutral memory layer and pricing depends on underlying Cloudflare resources |
No tagline | Agent Memory / Context Management | Cloud provider long-term memory | Microsoft Foundry Agent Service Memory | Azure public preview | Underlying model usage; memory pricing may change | Preview memory with underlying chat and embedding model billing | No separate free tier captured | Docs list maximum 100 scopes per memory store, 10,000 memories per scope, 1,000 search requests/min and 1,000 update requests/min | Long-term memory with user profile memory and chat summary memory; stores durable memory items in a managed memory store | Extraction, consolidation and retrieval phases manage preferences, summaries and conflicting facts | Designed for continuity across sessions, devices and workflows with personalized responses | Microsoft Foundry Agent Service, Memory Store API, Azure OpenAI chat and embedding deployments | Azure managed service | Preview terms apply; requires compatible Azure OpenAI chat and embedding model deployments | Azure subscription, scopes, IAM/project controls and Foundry Agent Service governance | Azure teams needing a managed long-term memory store tied to Foundry agents | Preview feature; automatic scope resolution is limited and pricing/billing can change during preview |
No tagline | Agent Memory / Context Management | Agent memory component | LlamaIndex Memory | MIT open-source framework feature | $0 software; model/storage costs separate | Framework memory abstraction | ✓ | No software cap; memory implementations depend on selected stores, LLMs and embeddings | Short-term FIFO message memory with optional long-term extracted memory; BaseMemory can be customized | Agent calls memory put/get or add/get; memory can trim to context window and optionally persist extracted information | Keeps agents within context limits and can add user/static blocks or long-term memory sources | LlamaIndex Python, LlamaIndexTS, agents, chat engines, memory blocks and adapters | Embedded in application code | Security depends on chosen storage, models and app controls | No separate team governance in the component | Existing LlamaIndex apps needing memory that fits the agent/query-engine ecosystem | Some older memory classes are deprecated or being replaced; production durability is not automatic |
No tagline | Agent Memory / Context Management | Long-term memory SDK | LangMem | MIT | $0 software | Open-source SDK; model, embedding and storage costs separate | ✓ | No software cap; production persistence requires a DB-backed LangGraph store such as AsyncPostgresStore | Structured long-term memories, hot-path memory tools and background memory managers | Extracts, consolidates, updates and searches memories using LangGraph BaseStore-compatible storage | Agents can manage memory during conversations or background processes can update it after interactions | LangGraph, LangChain, Python package, any BaseStore-compatible storage | Application code / self-hosted | Data path depends on model and storage providers; local stores can keep data local | No hosted team governance in the SDK itself | LangGraph teams needing first-party long-term memory primitives | Python-centric; in-memory examples are not durable; memory extraction can add LLM latency/cost |
No tagline | Agent Memory / Context Management | Memory management kit | ReMe | Open-source GitHub project | $0 software | Open-source memory management kit; model/storage costs separate | ✓ | No software cap captured in repository listing | Personal memory, task memory and tool memory for agents | Framework manages memory refinement and retrieval for agent workflows | Helps agents remember, refine and reuse information across interactions | AgentScope/modelscope ecosystem and Python agent workflows | Self-hosted/local application code | Data path depends on configured models and storage | No hosted governance captured | Developers who want a research/open-source kit for structured agent memory types | Official docs/pricing are GitHub-centric; production support and governance are limited |
No tagline | Agent Memory / Context Management | Local-first shared memory | Memobase | Local-first plus cloud sync | $0 Free; Pro $9/month; Unlimited $29/month | Credit-based monthly plans | ✓ | Free includes 500 credits/month; store_memory costs 10 credits and search costs 1 credit; Pro includes 5,000 credits/month; Unlimited lists unlimited credits | Persistent shared memory for AI agents with semantic vector search, per-user isolation and local SQLite mode | Local mode stores memories, entity relationships and architectural context in SQLite; cloud sync available with login | Adds cross-tool context continuity for Claude, ChatGPT, Claude Code and custom MCP agents | CLI, MCP server, Claude, Claude Desktop, ChatGPT and custom connector/API paths | 100% private local mode or managed cloud sync | Local mode sends no data to Memobase servers; cloud sync requires service trust | Free/Pro/Unlimited plan controls; dedicated support and SLA on Unlimited | Individuals and small teams wanting local-first shared memory for coding and chat agents | Two Memobase product lineages exist on the web; verify domain/product before adoption |
No tagline | Agent Memory / Context Management | Redis-backed memory service | Redis Agent Memory | Open source plus Redis Cloud service | $0 OSS; Redis Cloud pricing applies | Self-host Redis service or Redis Cloud private preview | ✓ | OSS server has no software cap; Redis Cloud service availability/pricing depends on account and preview access | Two-tier memory: session/working memory with TTL and long-term persistent memory | REST API and client libraries store, retrieve and manage contextual data in Redis with vector/keyword/hybrid search | Automatic lifecycle management, schemas and TTL keep agent memory structured and bounded | REST API, MCP, Python SDK, OpenAI, Anthropic and 100+ providers through LiteLLM | Self-hosted Redis Agent Memory Server or Redis Cloud | Self-hosting keeps data in customer Redis; Redis Cloud controls depend on plan | API key management and configurable schemas; broader governance follows Redis deployment | Teams already using Redis that want a production-ready memory service rather than custom tables | Managed cloud path is private preview; requires Redis/vector setup and model API keys for extraction |
No tagline | Agent Memory / Context Management | Framework persistence and memory | LangGraph Memory | Open-source framework feature / LangGraph Platform | $0 OSS; platform/infrastructure costs separate | Application-managed stores and checkpointers | ✓ | No OSS software cap; in-memory stores are ephemeral and production needs Postgres or another persistent backend | Short-term thread-level persistence and long-term user/application-level memory | Checkpointers save graph state; stores hold memories across threads and conversations | Manages conversation history, trimming and cross-session memories inside LangGraph agents | LangGraph, LangChain, Postgres stores, InMemoryStore and LangGraph Platform | Self-hosted app, LangGraph Platform or customer infrastructure | Security depends on selected database and platform; application owns scoping and retention | Governance depends on deployment and LangSmith/LangGraph Platform if used | LangGraph apps needing stateful workflows and memory without a separate SaaS | InMemorySaver loses data on restart; long-term memory design and scoping are still app responsibilities |
No tagline | Agent Memory / Context Management | Agent framework memory provider | Semantic Kernel Agent Memory | Experimental open-source framework feature | $0 software; Mem0/model/storage costs separate | AgentThread memory components and provider integrations | ✓ | No separate software cap; the documented Mem0Provider uses the Mem0 service and its pricing/quotas | Memories extracted from thread messages and supplied to agents; Mem0Provider enables cross-thread user memory | AgentThread components capture, retain and surface memory as needed | Allows agents to remember user preferences and context across multiple threads | Semantic Kernel Agents, Microsoft.SemanticKernel.Memory.Mem0Provider and vector stores | Application code / self-hosted with optional external memory service | Docs mark feature experimental; data path depends on Mem0 or configured memory provider | Governance follows app, provider and Azure/Microsoft stack if used | Semantic Kernel teams needing a Microsoft-native way to plug memory into AgentThread flows | Experimental and subject to change; current official guide centers on Mem0 integration rather than a standalone managed memory store |
No tagline | Agent Memory / Context Management | Agent SDK session memory | OpenAI Agents SDK Sessions | Open-source SDK feature | $0 SDK; API/model/storage costs separate | Built-in session memory with pluggable storage | ✓ | SDK has no software cap; local SQLite or hosted OpenAI Conversations storage can be used depending session implementation | Session memory stores conversation history for a specific session; sandbox agent memory is separate for reusable tips/preferences | Sessions automatically maintain conversation history across multiple agent runs | Eliminates manual to_input_list chaining and supports trimming/compaction patterns in SDK guides | OpenAI Agents SDK Python/JS, OpenAI ConversationsSession, SQLite/custom sessions | Application code plus optional OpenAI-hosted storage | Custom sessions can enforce retention policies, encryption and metadata; hosted storage follows OpenAI API policies | Governance depends on app storage implementation and OpenAI org controls | Developers building OpenAI Agents SDK apps that need short-term continuity and pluggable persistence | Session memory is primarily conversation history, not autonomous semantic/episodic memory unless paired with other tools |