HyperClaw v0.1.0 — Public Release

HyperClaw is live.

Your personal AI that actually gets things done.

What's New

• Turn-key onboarding for non-technical users
• Modern dashboard with Integration Hub (18+ services)
• One-click deploy (Railway, Render, Fly.io)
• 377 tests passing
• 50+ specialized agents
• 30+ platform integrations

Quick Start

git clone https://github.com/mentatalbans/hyperclaw.git
cd hyperclaw
pip install -e .
hyperclaw init

Requirements

• Python 3.11+
• Anthropic API key

Integrations

Telegram, Slack, Discord, WhatsApp, Gmail, Google Calendar, Notion, Salesforce, HubSpot, GitHub, Jira, Stripe, and more.

---

Built by GIL.

⚡ HyperClaw v0.1.0-alpha

HyperClaw v0.1.0-alpha — First Release

The AI that actually takes over. 22 specialist agents. UCB1 bandit routing. Empirical code certification. Recursive self-improvement.

What's in this release

HyperCore

• HyperState — Pydantic v2 state management with domain, task, hypothesis, experiment, and certification tracking
• HyperRouter — UCB1 multi-armed bandit model/agent selection (gets smarter with every task)
• FastLoop + SlowLoop — real-time routing + background weight optimization
• Certifier — empirical method certification (test_trace required, empty results blocked)

Multi-Model Stack

• ClaudeClient — async Anthropic with 3x retry + streaming
• ChatJimmyClient — Taalas HC1 / Llama 3.1 8B @ 17,000 tok/sec (never auto-certified)
• ClaudeCodeSubagent — generate → run → fix → certify loop (up to 5 iterations)
• ModelRouter — UCB1-driven dispatch + SwarmMessage logging

HyperMemory

• CausalGraph — PostgreSQL + pgvector knowledge graph. Every certified method writes cause→effect.
• VectorStore — cosine similarity search with domain/type filters
• ImpactTracker — measures and surfaces business/scientific impact by domain
• AgentMemory — per-agent episodic, semantic, and procedural memory

HyperShield

• PolicyEngine — YAML policies with hot-reload, per-agent network/filesystem isolation
• NetworkGuard — egress allowlists, raises NetworkEgressBlockedError on violations
• FilesystemGuard — sandbox enforcement, raises FilesystemAccessBlockedError
• AuditLogger — immutable DB-backed audit trail for all agent actions

HyperSwarm — 22 Specialist Agents

• Personal: ATLAS, MIDAS, VITALS, NOURISH, NAVIGATOR, HEARTH
• Business: STRATEGOS, HERALD, PIPELINE, LEDGER, COUNSEL, TALENT
• Scientific: MEDICUS, COSMOS, GAIA, ORACLE, SCRIBE
• Creative: AUTHOR, LENS
• Recursive: SCOUT, ALCHEMIST, CALIBRATOR
• Orchestrator: NEXUS (UCB1 bid protocol, HERALD assembly)

Recursive Growth Engine

• SCOUT → sweeps arXiv, GitHub, PubMed every 6 hours
• ALCHEMIST → validates discoveries, implements via ClaudeCodeSubagent, certifies new skills
• CALIBRATOR → reads SwarmMessage log, surfaces routing inefficiencies

TUI

• Full Textual TUI: Dashboard, HyperState, Knowledge, Impact, Research, Audit
• Keyboard shortcuts: d/s/k/i/r/a to switch screens, q to quit

Install

pip install hyperclaw
hyperclaw init
hyperclaw start

Tests

184/184 passing. Run with:

pytest tests/ -v --cov=core --cov=memory --cov=security --cov=models

Requirements

• Python 3.11+
• PostgreSQL with pgvector (for HyperMemory — optional for basic swarm use)
• ANTHROPIC_API_KEY environment variable

Notes

• This is an alpha release. APIs may change before v0.2.0.
• ChatJimmy integration requires a Taalas API key (free tier available at chatjimmy.ai)
• pgvector-dependent features (VectorStore, CausalGraph, AgentMemory) require a live PostgreSQL connection