🐝 GEIANT Hive Provider for OpenClaw

Cryptographically-signed local AI inference. Zero cloud. Zero data leakage. EU AI Act ready.

This plugin gives OpenClaw two things at once:

1. Distributed local inference — replaces cloud LLM APIs (Anthropic, OpenAI) with the GEIANT Hive swarm. Inference runs on devices you control; data never leaves your network.

2. Cryptographic agent identity — every OpenClaw decision, tool call, and inference becomes a signed breadcrumb in the GNS Protocol audit chain. Verifiable. Jurisdictional. EU AI Act Article 12 compliant by architecture.

Status: Sprint 2 alpha — initial scaffolding shipped, identity layer and signing in progress. See Roadmap.

Why?

| | Cloud APIs (Anthropic / OpenAI) | GEIANT Hive | |---|---|---| | Data residency | Sent to US/EU cloud servers | Never leaves your network | | Cost | $0.002 – $0.06 per token | Zero per-token (your hardware) | | Audit trail | Trust the provider | Cryptographically verifiable | | Agent identity | None | Ed25519 keypair, signed actions | | Jurisdiction proof | None | H3 cell bound, EU AI Act ready | | 500 users | $10–30K/month | €990/month flat | | Offline | No | Yes |

Architecture

The plugin owns a GNS identity. Every request flows through it:

OpenClaw process starts
  ↓
@gns-foundation/openclaw-hive-provider loads
  ↓
Plugin reads or generates Ed25519 keypair
  (at $GEIANT_AGENT_IDENTITY_PATH, default
   ~/.config/geiant-hive/identity.json)
  ↓
Plugin registers identity with the swarm
  ↓
[OpenClaw normal operation]
  ↓
When OpenClaw calls the LLM provider:
  Plugin signs request with the agent's Ed25519 key
  POSTs to https://hive.geiant.com/v1/chat/completions
  with X-GNS-PublicKey, X-GNS-Timestamp, X-GNS-Signature
  ↓
Backend verifies signature → routes to swarm → returns response
  Audit trail: every chat completion is cryptographically tied
  to the agent's GNS identity, queryable forever.

This is what makes GEIANT a credible LangChain / CrewAI / AutoGen alternative for regulated and sovereign workloads: not by winning on ecosystem (we won't), but by covering the same functional ground with structurally better properties — verifiable audit, jurisdictional routing, cryptographic accountability.

Roadmap

This repo is the Sprint 2 work item from the GEIANT Hive Roadmap §12.3. Progress:

• ✅ Phase 0 — repo scaffolding (this commit)
• ⏳ Phase 1 — backend identity layer (in gns-backend repo)
  • Migration 008: agents table
  • POST /v1/agents/register endpoint
  • signed_request.ts middleware extension to recognize agent identities
• ⏳ Phase 2 — plugin implementation (this repo)
  • Identity bootstrap (src/identity.ts) — load or generate Ed25519 keypair
  • Signing module (src/sign.ts) — using tweetnacl, matches the worker's pattern
  • Registration flow (src/register.ts) — first-run, POSTs to /v1/agents/register
  • Request interceptor — every outbound /v1/chat/completions gets signed
• ⏳ Phase 3 — demo artifact + publish
  • npm publish 0.1.0-alpha.1
  • README screenshots
  • Live OpenClaw session with visible audit trail at hive.geiant.com/audit

Quick Setup (will work after Phase 2 ships)

# Install
npm install @gns-foundation/openclaw-hive-provider

# Configure OpenClaw to use Hive
openclaw onboard --provider geiant-hive --hive-url https://hive.geiant.com/v1

# First run: plugin generates a GNS keypair and registers as an agent.
# Subsequent runs: signs every inference call with the agent's identity.

Available Models (in production today)

| Model ref | Provider | Parameters | Best for | |---|---|---|---| | geiant-hive/lfm2.5-1.2b-instruct | Liquid AI | 1.2B | Default — fast, capable, edge-optimized | | geiant-hive/lfm2.5-1.2b-thinking | Liquid AI | 1.2B | Multi-step reasoning, slower per token | | geiant-hive/phi-3-mini | Microsoft | 3.8B | Strong general-purpose | | geiant-hive/tinyllama | TinyLlama Project | 1.1B | Smoke tests, low-resource fallback |

More models planned for v0.7+ (Llama 3.1 8B, Mistral 7B, larger Liquid AI models).

Why "Liquid AI" prominently?

We run Liquid AI's LFM2.5 family as the default Hive model. LFM2.5 is a frontier-quality small model with state-space architecture, well-suited for edge deployment on a decentralized swarm. Crediting upstream is good citizenship and good signal.

Configuration

| Setting | Env var | Default | |---|---|---| | Identity file path | GEIANT_AGENT_IDENTITY_PATH | ~/.config/geiant-hive/identity.json | | Hive gateway URL | HIVE_URL | https://hive.geiant.com/v1 | | Jurisdiction (H3 cell) | (auto-detected) | from IP / timezone | | Min worker trust tier | (per-call header) | seedling (accept all) |

License

Apache-2.0. Same permissive license as the rest of the @gns-foundation ecosystem and OpenClaw itself.

ULISSY s.r.l. · Via Gaetano Sacchi 16, 00153 Roma · GNS Protocol · Proof-of-Trajectory™

"OpenClaw gave AI agents hands. GEIANT gives them an identity, a jurisdiction, and a wallet."