I am building TMCRA, a self-hosted temporal-memory platform for AI agents and products.
TMCRA recalls owner-global and current-project evidence before an agent responds, then stores the user message and the agent result as separate, source-attributed records. Across supported tools and conversations, it preserves project decisions, requirements, progress, test results, unresolved problems, and next steps. Accumulated collaboration can be organized into visual memory graphs and personal knowledge.
我在开发 TMCRA：一个可自托管的 Agent 与产品时序记忆平台。它让 Agent 在不同对话和工具之间延续同一项目，并将长期协作中的需求、决策、进度、验证结果与经验沉淀为可追溯的记忆和个人知识。
Website · Open-source platform · Core runtime · Community discussions
| Repository | Category | Purpose | Status |
|---|---|---|---|
| tmcra | Platform | Full self-hosted source release: API, web, desktop, mobile, SDK integrations, Codex plugins, MCP, benchmarks, and assets | Stable |
| tmcra-memory | Core | Local runtime, scoped memory, evidence retrieval, visual memory graph, personal knowledge, and benchmark package | Active |
| tmcra-plugin-codex | Integration | Automatic recall and role-separated writeback for Codex lifecycle hooks | Active |
| tmcra-plugin-deepseek-harness | Integration | Automatic project memory for DeepSeek Harness | Technical preview |
| tmcra-mcp-server | Protocol | MCP server with explicit recall, prepare, commit, job, and reconciliation tools | Active |
| Repository | Focus | Status |
|---|---|---|
| tmcra-token-graph | Graph-native autoregressive language-model research without dense causal self-attention | Experimental |
For integration questions, bug reports, and technical discussion, use TMCRA Discussions.