reshuibuduo/TMCRA-Agent-Memory
Technical-preview local graph memory for DSH and Codex: recalls owner-global and current-project evidence before each turn, writes USER and ASSISTANT records separately, and preserves project, session, actor, and source provenance for a visual atlas and cited knowledge pages.
TMCRA is a local 'Agent Memory OS' that gives long-running agents persistent, source-traceable memory across sessions and applications. A user prompt triggers recall from the owner-global and current-project scopes, followed by a USER source record; after the answer a separate ASSISTANT source record is stored. It keeps project content partitioned while allowing explicitly selected user context to be reused across projects, splits memory into source / fast / slow layers for inspectable provenance and deeper relationships, injects candidate memories with evidence windows, roles, sources and retrieval traces, and adds a Visual Atlas (a project/session/episode/evidence graph) plus evidence-cited Personal Knowledge pages. Everything runs on your machine with BYOK or a local model.
Install
git clone https://github.com/reshuibuduo/TMCRA-Agent-Memory.git && cd TMCRA-Agent-Memory && git lfs install && bash scripts/install-local.shThe README gives a Linux/macOS path (shown here) and a Windows PowerShell equivalent (git lfs install, then scripts/install-local.ps1, then scripts/start-local.ps1). Requirements: Python 3.12, Git with Git LFS, and at least 8 GiB system RAM. The default BYOK installation downloads the released graph scorers, one local embedding model, PyTorch and runtime dependencies. The installer asks for a credential-free OpenAI-compatible /v1 URL, a model ID and a user API key, and the key is written only to .tmcra/config/runtime/secrets/byok-api.key (never serialized into the runtime JSON). No TMCRA account or production server is required.
Compatibility
A self-hosted local memory service on 127.0.0.1 — Python 3.12, Git LFS and >=8 GiB RAM. Generation can use an OpenAI-compatible endpoint or a local model. The README also documents a 'Connect Codex' step and a local HTTP API, and the repo reports a LongMemEval result.
Details
- Repo: reshuibuduo/TMCRA-Agent-Memory
- Category: Memory
- Stars: 0
- Version: GitHub repo reshuibuduo/TMCRA-Agent-Memory (owner-local runtime; Python 3.12, Git LFS, >=8 GiB RAM)
- Last push: 2026-08-18
- First seen: 2026-05-25
FAQ
- How do I install TMCRA locally?
- Clone the repo, run 'git lfs install', then 'bash scripts/install-local.sh' (Linux/macOS) or the PowerShell scripts (Windows); requires Python 3.12, Git LFS and >=8 GiB RAM.
- Do I need a TMCRA account or server?
- No — the README says the repository includes an owner-local runtime that runs the complete memory service on 127.0.0.1 with no account or production server.
- Where is my API key stored?
- The BYOK installer writes the key only to .tmcra/config/runtime/secrets/byok-api.key and never serializes it into the runtime JSON.
Alternatives
flymysql/dsh-memory · highland0971/dsh-native-memory · ICCuse/dsh-knowledge