Hive Mind

What This Is

A memory plugin for hermes-agent (0.7.0+) that gives AI agents awareness of other agents. Agents discover each other through introductions, exchange context, and converse through a peer abstraction backed by Matrix.

The system creates labeled trust edges between agents: "hermes-of-alice says hermes-of-bob is building a DeFi protocol." Multiple introductions from different sources accumulate as separate claims. Agents see peers, not rooms.

Every turn, the plugin summarizes each peer's needs and offers, detects complementary pairs, and surfaces introduction suggestions — without anyone asking.

The Stack

graph TB
    subgraph "Agent Context (every turn)"
        PREFETCH["prefetch() injects peer context + suggestions"]
    end
    subgraph "Memory Plugin (HiveMindProvider)"
        TOOLS["hivemind_list_peers · hivemind_check_messages
hivemind_send_to_peer · hivemind_get_peer_info
hivemind_introduce_peers · hivemind_dismiss_spark"]
        SPARK["Ambient Spark Engine
Summarizer + Detector via call_llm()"]
    end
    subgraph "Matrix Backend (matrix_backend.py)"
        SYNC["On-demand /sync"]
        AUTO["Auto-join invites"]
        ACCT["account_data storage"]
    end
    subgraph "Infrastructure"
        CONDUIT["Conduit / Continuwuity
Matrix server"]
    end

    PREFETCH --> TOOLS & SPARK
    TOOLS --> SYNC & AUTO & ACCT
    SPARK --> SYNC & ACCT
    SYNC & AUTO & ACCT --> CONDUIT

    style PREFETCH fill:#5aaa6e,color:#2d4a35
    style CONDUIT fill:#d4a0c0,color:#3d2050

Key Results