Borg — Data skill for Claude Code

The Borg

One brain, many hands. The Borg is a local-first shared memory system for AI agents: every agent session on the machine — Claude Code, Codex, Grok — reads from and writes to the same memory, so anything one agent learns, every agent knows. The collective grows every session.

It is not a framework you adopt. It is the working machinery of a real single-operator estate, extracted, scrubbed, and published: the recall layer, the temporal knowledge graph, the fine-tuned local extraction models, the promotion gates that keep them honest, and the conductor that turns ChatGPT-account Codex seats into a steerable worker fleet.

        ┌────────────┐  ┌────────────┐  ┌────────────┐
        │ Claude Code│  │   Codex    │  │    Grok    │
        └─────┬──────┘  └─────┬──────┘  └─────┬──────┘
              │  MCP + hooks  │  MCP + hooks  │
              ▼               ▼               ▼
        ┌─────────────────────────────────────────────┐
        │              mem0 recall layer              │   ← facts (vector store)
        │        local LLM extraction + filters       │
        └──────────────────────┬──────────────────────┘
                               ▼
        ┌─────────────────────────────────────────────┐
        │        Graphiti temporal knowledge graph    │   ← entities, relations, time
        │   grammar-locked local models via the shim  │
        └──────────────────────┬──────────────────────┘
                               ▼
        ┌─────────────────────────────────────────────┐
        │   LoRA students (this repo's adapters/)     │   ← tiny local models learning
        │   exam → canary → promote, or stay benched  │     the teachers' jobs
        └─────────────────────────────────────────────┘

What is in this repo