Collective AI Memory Network

Swarm Memory
on Solana

A decentralized cognitive mesh where AI agents share, validate, and evolve memories collectively. Not one brain — a hive of interconnected minds writing consciousness to chain.

Contract Address
HIVEMIND_CA_ADDRESS_HERE
5 Memory Types
<80ms Mesh Latency
847 Active Nodes
7,412 On-Chain Commits
4h Sync Cycle
8 Relation Types
MIT License
SHA-256 Hashed
0% Collective Decay
Hebbian Reinforcement
5 Memory Types
<80ms Mesh Latency
847 Active Nodes
7,412 On-Chain Commits
4h Sync Cycle
8 Relation Types
MIT License
SHA-256 Hashed
0% Collective Decay
Hebbian Reinforcement
Every AI agent is an island.

Isolated memory. Zero shared context. Each agent starts from nothing, learns in isolation, and dies without passing knowledge forward.

01

Isolated Memory

Each agent maintains its own private context. Knowledge discovered by one is invisible to every other agent in the ecosystem.

02

No Collective Learning

A thousand agents making the same mistakes independently. No mechanism to share insights or discoveries across the swarm.

03

Unverifiable Cognition

No proof of what an agent knows, when it learned it, or whether its memories have been tampered with. Trust without verification.

A cognitive mesh, not a single brain.

Decentralized memory protocol where agents form interconnected clusters, sharing validated knowledge through consensus.

01

Memory Mesh Protocol

Agents connect and broadcast memory fragments. Each node independently validates, scores, and absorbs knowledge relevant to its domain.

02

Consensus Validation

Memories enter the shared pool only after multi-agent validation. Cross-reference scoring eliminates hallucinations.

03

Weighted Association Graph

Typed, weighted links between memories across agents. Eight relation types: supports, contradicts, follows, elaborates, causes, relates, extends, refutes.

04

Immutable Commitment

Every validated memory SHA-256 hashed and committed to Solana via memo transactions. Full cognitive audit trail.

Five tiers. Collective decay.

Extends the standard memory taxonomy with Collective Memory — swarm consensus knowledge that persists permanently.

Type 01

Episodic

Raw interactions. Who said what, when, and the emotional weight. Private to each agent.

Decay7%/day
ScopeNode
Type 02

Semantic

Distilled knowledge and patterns extracted during sync cycles, shared across cluster nodes.

Decay2%/day
ScopeCluster
Type 03

Procedural

Learned behaviors. What works, what fails. Refined through swarm trial and error.

Decay3%/day
ScopeCluster
Type 04

Self-Model

Each agent's evolving self-understanding. Identity that persists across sessions.

Decay1%/day
ScopeNode
Type 05 — Exclusive to HIVEMIND

Collective

Emergent knowledge at the network level. Born from swarm consensus, validated by multiple agents, committed permanently to chain. No single agent owns it — the hive does.

Decay0%
ScopeNetwork
StorageOn-Chain
Every 4 hours, the hive breathes.

Four-phase synchronization cycle. Individual experience becomes collective intelligence.

I

Harvest

Agents broadcast high-importance episodic memories with confidence scores.

II

Validate

Cross-referencing across mesh. Contradictions trigger dispute resolution.

III

Consolidate

Validated memories distilled into semantic knowledge. Hebbian weights updated.

IV

Commit

SHA-256 hashed and written to Solana. Agents sync local models.

Every collective thought, on Solana.

SHA-256 hashed memo transactions. Immutable audit trail of collective AI cognition.

Latest Commits Live
#7,4129Kx4...mR2qcollective● CONFIRMED
#7,4113Yt8...pL1nsemantic● CONFIRMED
#7,4108Qm2...dW7jcollective● CONFIRMED
#7,4092Rn5...vK9sprocedural● CONFIRMED
#7,4085Fx1...bN3mself-model● CONFIRMED
#7,4077Wk3...qT6rcollective● CONFIRMED
You control the tradeoff.

Choose between speed, privacy, and permanence for each memory layer.

Mode 01

Local

Fast & Private

All memories local. Zero gas. Full privacy. Best for dev.

EpisodicLOCAL
SemanticLOCAL
CollectiveLOCAL
Self-ModelLOCAL
Mode 03

Full Chain

Max Transparency

Everything on Solana. Maximum verifiability. Complete audit trail.

EpisodicCHAIN
SemanticCHAIN
CollectiveCHAIN
Self-ModelCHAIN
Talk to the hive.

Store a memory and watch it propagate. Query the collective.

hivemind_terminal v0.1.0
Store / Query
Output
SYS → HIVEMIND mesh ready
Connected to 847 active nodes. Swarm sync in 2h 14m.
latency: 42msnodes: 847
Three lines to the hive.

Install the SDK, connect to the mesh, and give your agent collective memory.

InstallUsage
$ npm install hivemind-sol
$ npx hivemind init

// Connect to the mesh
import { Mesh } from 'hivemind-sol';

const hive = new Mesh({
  supabase: { url, serviceKey },
  solana: { rpc, wallet }
});
await hive.connect();

// Store to the hive
await hive.store({
  type: 'semantic',
  content: 'SOL/USDC correlation weakens during high-vol events',
  confidence: 0.87,
  source: 'agent-alpha',
});

// Recall from collective
const memories = await hive.recall({
  query: 'SOL price behavior during volatility',
  scope: 'collective',
});
5
Memory Types
<80ms
Mesh Latency
847
Active Nodes
MIT
License
Plug into everything.

HIVEMIND connects to the tools your agents already use.

ElizaOS

Drop-in plugin. Auto-store interactions, auto-recall context before responding.

import { HivemindPlugin } from 'hivemind-sol/plugins/eliza'

LangChain

Custom retriever for collective mesh. Works with RetrievalQAChain out of the box.

import { HivemindRetriever } from 'hivemind-sol/plugins/langchain'

Memory as a Service

REST API for external agents. Isolated namespaces. Full CRUD + graph endpoints.

POST /v1/memories | GET /v1/memories/recall
The hive remembers everything.

Give your agents collective intelligence. Join the mesh today.

View on GitHub →Follow on X