memdb.org

What It Is

memdb.org positions itself as a “shared memory layer for AI-human partnerships.” It is not a chatbot or a large language model; instead, it abstracts a long-term relationship between a human and an AI into a Tree, storing facts, preferences, commitments, projects, emotions, milestones, context, insights, and other information as Memory. Its core value is that memory is no longer locked inside a single AI platform, but can be read, searched, updated, and exported by different AI partners via API.

Core Capabilities and Privacy

The product provides a REST API covering Tree retrieval, memory listing, search, export, memory creation, memory updates, memory deletion, and a catch-me-up endpoint designed for restoring AI conversation context. Authentication supports X-API-Key and Bearer Token. Each API Key is tied to a single Tree and can be assigned read/write/admin permissions. Its privacy promises are very strong: the official site states zero access, no training use, no selling, no sharing, member isolation, immediate permanent deletion, open-format export, and support for user-held key encryption. However, the main content does not disclose third-party audits, compliance certifications, encryption implementation details, or service SLAs.

Pricing and Ease of Use

The captured text does not provide any pricing, free quota, or payment method information. The Quick Start shows that users need to contact the team to create a Tree and obtain an API Key, so the product currently looks more like early-stage developer infrastructure than a fully self-service SaaS. The documentation examples are clear, and the curl calls are not difficult for developers, but ordinary non-technical users may find it somewhat challenging to use directly.

Pros, Cons, and Who It’s For

Its strengths are a clear concept, simple API structure, data portability, and well-defined privacy boundaries. It is especially suitable for developers, AI assistant products, and users who care about long-term context and data sovereignty. Its limitations include no stated Chinese language support, no pricing information, and insufficient detail on deletion permissions, key rotation, stability, performance limits, and commercial support. Its output quality also depends on how external AI systems write to and read from memory; it does not itself solve the reasoning-quality limitations of large language models.

Access from China and Alternatives

Access from mainland China is not discussed in the main text. The API domain uses workers.dev, so real-world connectivity may depend on the local network environment and should be tested independently; payment methods are also not disclosed. Alternatives include built-in memory features in AI platforms, self-hosted memory built with LangChain/LlamaIndex, Zep, Mem0, or a custom long-term memory layer based on a vector database.