AI Agent Hub
Back to plugins
memobranch preview

memobranch

Memory Updated 2026.09.04

Run the following command in DeepSeek Harness:

dsh plugin install sens-io/memobranch

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install sens-io/memobranch inside DeepSeek Harness to install this plugin; the source repository is at https://github.com/sens-io/memobranch

About this plugin

Most AI agent memories are nothing but a vector store: where the content came from, who modified it, and how conflicting updates were resolved are questions that simply go unanswered. MemoBranch redefines long-term memory as a governable knowledge chain — Markdown is the source of truth, Git records every change with authorship, and the LLM is an optional enhancement layer only. Even when the model API is down, capture, review, CJK/English lexical search, and remote sync all keep working.

Knowledge flows through a three-stage pipeline: evidence (append-only raw facts) to candidates (refined and queued for review) to wiki (formal, searchable memories). Server-side access control checks permission, scope, sensitivity, and tenant before any decryption occurs. Sensitive records are protected with per-record AES-256-GCM envelope encryption; logical keys use opaque paths and are excluded from Git indexes and vector APIs. Multi-file writes go through a transaction journal followed by atomic replacement, enabling precise rollback or full replay after a crash. Remote sync is built on standard Git with ahead/behind/diverged detection and conflict abort, delegating authentication to credential helpers so tokens never touch disk.

MemoBranch plugs into agent toolchains as either a native DeepSeek Harness plugin or an MCP toolset. It is well suited for teams and developers who need auditable, shareable long-term memory across multiple agent instances, who prioritise data sovereignty and compliance, and who run in local or private-cloud environments.

Screenshots

Use Cases

  • Sharing auditable, traceable long-term knowledge across multiple agent instances
  • Production systems requiring provenance tracking, revision history, and conflict handling
  • Offline or degraded environments where CJK/English lexical search and Git sync must still work without a model API

Best For

  • Engineering teams building multi-agent systems with auditable knowledge requirements
  • Developers and platform teams working within the DeepSeek Harness or MCP toolchain
  • Enterprise users with strict data-sovereignty, compliance, and local-first deployment requirements