MAXIM
Federated Bio-Substrate Layer for Collective Cognition
Design complete (2026-05-09). Implementation phased: shareability infrastructure in v1.0 (B5), Oasis software in v1.1, full Hivemind P2P protocol in v1.2. Supersedes the older Mother Maxim plan (2,224 LOC) with a more bio-coherent ~2,100 LOC architecture that's federated, not hierarchical.
The Maxim Hivemind is a peer-to-peer substrate-sharing layer that lets multiple Maxim instances exchange learned bio-substrate (NAc causal weights, EC concept centroids, reflex thresholds, Cerebellum forward models, ATL semantic concepts). It is not a service — nobody hosts "the Hivemind." It is a protocol + portable data format + exchange convention.
A Maxim Oasis is a persistent, substrate-primary Maxim instance that runs on operator hardware (a Mac Mini, a server, an old laptop, a Raspberry Pi 5 cluster), accepts substrate contributions from connected Maxims, processes those contributions as percepts through her own bio-stack, distills consensus patterns, and broadcasts to the Hivemind. Multiple Oases coexist; nobody is THE Oasis. Each Oasis is a real bio-agent in her own right, not a passive database.
The headline insight that motivated promoting the old Mother Maxim plan from deferred to active:
Every LLM-AUT run already trains a bio-substrate as a side effect. When a user runs maxim --sim "...some D&D campaign...", the bio-systems are running underneath: NAc forms causal links from tool outcomes, EC clusters concepts, ATL builds semantic structure, Hippocampus captures episodes. The substrate is being trained right now by your existing user base, just by an LLM-driven action selector instead of a substrate-driven one.
That changes the federated-learning calculus radically:
It doesn't have to wait for substrate-primary mode to mature. Day-1 contributions come from existing LLM-AUT users.
When substrate-primary mode lands in v1.1, new instances can pull a baseline substrate from the Hivemind rather than starting from zero. Massively accelerates substrate-primary development.
Distilled bio-substrate doesn't carry the same PII risk as raw episodes. The 700-LOC dual-pass deidentification pipeline from the old plan becomes optional polish rather than load-bearing safety.
It's not transitional waiting to be replaced; it's the perpetual data-collection mechanism that feeds the Hivemind. Users running D&D campaigns today are inadvertently training tomorrow's substrate-primary cognition.
| Substrate component | Shareability | Privacy risk | Hivemind value |
|---|---|---|---|
| Reflex specs + thresholds | Trivial — pure config | Zero | High |
| NAc causal weights | High — Bayesian aggregation | Low | Very high |
| EC concept centroids | Medium — encoder-dependent | Low (after identity scrub) | High |
| Cerebellum forward models | Medium — parameter snapshots | Zero | Medium |
| ATL semantic concepts | Medium — cluster purity | Low | Medium |
| Hippocampus episodes | LOW — literal records | HIGH | Low |
| Working memory | None — ephemeral | High | Zero |
Default policy: ship everything in the top section. NEVER ship hippocampus episodes through the Hivemind. Episodes are local-only by construction. The single exception: an Oasis may receive hippocampus episodes from connected Maxims as private contributions to her own substrate (she processes them as percepts), but she never re-broadcasts them. Episodes feed the Oasis's substrate; only the Oasis's distilled substrate ever leaves.
A bio-coherent metaphor: an oasis is a sustaining gathering place that travelers approach to refresh themselves and that produces life from accumulated water/nutrients. Maxims approach an Oasis to contribute experience and to receive distilled patterns. Multiple oases coexist in a desert ecosystem; no single one is canonical; together they form a network.
A Mac Mini, an old laptop, or a Raspberry Pi 5 cluster is sufficient to run an Oasis IF substrate-primary mode is the operating mode (no LLM, just bio-systems running). During the transitional phase (v1.1) when substrate-primary is still maturing, Oases may need a small LLM (e.g., qwen-7B) to process LLM-AUT contributions. Once substrate-primary works (post-Phase 0/1), Oases can run pure-substrate and Mac-Mini-class hardware suffices.
There is no THE Oasis. Each operator runs their own. A natural evolution emerges:
Your Mac Mini, hosting substrate from your own Maxims and friends.
A Discord server or research group running an Oasis their members contribute to.
Eventual reference instances anyone can connect to.
The unit of exchange. A versioned, signed archive containing:
maxim-substrate.zip
├── manifest.json # _format_version, schema_version, contributor_id, domain, signature slots
├── nac.json # causal links + confidence + provenance tags (v1.0)
└── ec.json # concept centroids + cluster metadata (v1.0)ATL, reflexes, and Cerebellum payloads are reserved for v1.1 (the migration-registry seam is already in place). Schema-stable, version-aware, scrubbable. Extends the existing _format_version contract per Maxim's v1.0 freeze.
Bundles are tagged by substrate domain (combat, cooking, medical, fantasy, robotics, conversation, ...). Subscribers opt into specific domains. Two purposes: relevance (a medical-AI hobbyist doesn't want combat patterns polluting their substrate) and curation (domain maintainers can flag bad patterns).
Two Maxims learn opposite things ("X is good" vs "X is bad"). On merge, the Hivemind uses NAc's existing confidence math:
When real conflict persists (e.g., 50 Maxims learned each side), the Hivemind preserves both with full distributions. The local Maxim's behavior remains coherent because it weights its own learning above any single Hivemind pattern.
Adversaries trying to corrupt the Hivemind by spreading bad learning are mitigated by:
This won't be perfect — pure peer-to-peer systems with open contribution always have an arms race with abuse. Mitigations are practical, not theoretical.
| Version | Substrate-primary | Oasis | Hivemind |
|---|---|---|---|
| v1.0 | B5: Phase −1 + Phase 0 harness + shareability infra | Format exists. No software runs. | None. |
| v1.1 | Substrate-primary AUT mode lands. Phase 0 validation. Phase 1 starts. | Oasis software ships (~800 LOC). Mac-Mini hostable. maxim oasis serve. |
Direct Oasis-to-Oasis sync only. |
| v1.2 | Phase 1 ships. Substrate-primary Maxims pull bootstrap. | Multi-Oasis federation. Curation tools. | Full Hivemind P2P protocol (~600 LOC). |
| v1.3+ | Phase 3 from-scratch sequence model. | Pure substrate-primary; no LLM needed. | Cross-version migration; domain ecosystem. |
The 2,224-line Mother Maxim plan was designed in the LLM-AUT-only world where memories (dialogue + episode traces) were the unit of value. The 2026-05-09 substrate-primary pivot reframed the architecture significantly.
/v1/contribute → Hivemind P2P protocol