sens-io/memobranch

Auditable, Git-native long-term memory for AI agents: conversational evidence, candidate knowledge and canonical memories compiled into a human-readable Markdown Wiki, working fully without a model API.

MemoBranch is a local-first long-term memory layer that organizes conversational evidence, candidate knowledge and canonical memories into a human-readable Markdown Wiki, with Git providing versioning, attribution, rollback and cross-machine synchronization. Inspired by OpenKnowledge's Git + LLM Wiki approach (an independent implementation, per the README), it follows Karpathy's method — continuously compiling and maintaining interconnected, source-backed knowledge rather than only vector search — with incremental cross-source compilation, explicit answer filing and semantic Lint. It ships a web management console and an MCP integration alongside the DSH plugin, runs under the OpenSpec proposal -> specs -> design -> tasks -> implementation -> verification workflow, and pairs locally verified acceptance with an isolated independent review record. The README is careful to note that local acceptance does not establish GitHub-hosted CI results, real-model quality, or production release validation.

Memory ★ 2
View on GitHub ↗

Install

dsh plugin --profile personal-agent add memobranch

Copied verbatim from the README's DeepSeek Harness plugin section (into a personal-agent profile). The README also documents a local-path form and a GitHub form (dsh plugin --profile personal-agent add github:sens-io/memobranch#<commit-sha>).

Compatibility

A DSH plugin (also MCP-ready). Local-first and Markdown/Git-native. The README stresses that an LLM is not a source of truth: capture, review, Git versioning, recovery, Chinese/English retrieval, and the DeepSeek Harness and MCP integration all remain fully functional without a model API.

Details

Alternatives

omdsh-dev/dsh-mnemon · diqierjia/StrataGate-AgentMemory · kanhaoning/NeatMem

More plugins in Memory

Browse more in Memory

Guides for Memory plugins