tribalmemory.com

What it is

TribalMemory is a shared memory store for AI agents, with the core tagline “One memory store. Every agent connected.” It addresses the problem of multiple coding agents forgetting information independently and lacking shared context: once one agent saves project knowledge, other agents can retrieve it from the shared memory. For example, Claude Code might record that authentication uses JWT RS256, and Codex CLI can later answer questions based on that memory.

Core capabilities

Based on the page content, TribalMemory focuses on retrieval augmentation and agent integration. It supports semantic search, graph search, hybrid vector and BM25 retrieval, entity-relationship traversal, session indexing, deduplication, temporal reasoning, and JSON import/export. By default, it uses FastEmbed ONNX to compute embeddings locally, with memories stored in LanceDB. Users can also choose OpenAI embeddings or Ollama. It also offers native MCP integration, works with Claude Code and Codex CLI, and provides an example OpenClaw plugin.

Pricing and privacy

The page does not disclose commercial pricing or paid plans. It shows installation via pip install, lists the version as PyPI v0.7.0, and notes a BSL 1.1 license. Its privacy positioning is clear: in local mode, embeddings are computed locally and data is stored locally, with no API key, no cloud, and no telemetry. This makes it suitable for users who are sensitive about sending code and project knowledge outside their environment. However, if OpenAI embeddings are selected, external API usage is involved and should be evaluated separately.

Pros, cons, and limitations

Its strengths are a clear installation path—tribalmemory init and serve are enough to get started—plus a local-first approach that reduces privacy and cost concerns. MCP also allows it to fit naturally into mainstream agent-based coding tools. The limitations are that the page does not provide information on Chinese-language support, enterprise deployment, access control, backup/sync, SLA, or commercial support. The showcased metrics also include items such as “0 tests passing” and “0 LoCoMo recall,” so real-world stability and retrieval quality still need testing. The BSL 1.1 license may also affect some commercial use cases.

Who it’s for and access from China

TribalMemory is best suited to developers who frequently use tools such as Claude Code, Codex CLI, and OpenClaw, especially teams that want multiple agents to share project conventions, architectural decisions, and historical sessions. The page does not make China accessibility clear. Since it can be installed locally, its core runtime theoretically does not depend on the cloud, but resources such as GitHub, PyPI, and Discord may be unstable on mainland Chinese networks. Payment information is not disclosed. Alternatives include building a vector database plus MCP setup yourself, using built-in memory features in AI IDEs, or adopting local knowledge base tools.