VaHive Systems Lab

DSMC / MAGUS

Governed cognitive architectures for long-lived AI systems. Context drift is not a model failure — it is a memory architecture failure. These are the structural tools to fix it.

↗ vahive.co ↗ manifest.vahive.co ✉ [email protected]

Start here

Free — working code + architecture docs

Token Efficiency Suite — free working code

dsmc_minimal.py + dsmc_minimal_sidecar.py + TOKEN_EFFICIENCY.md — zero-dependency Python active state governance. No pip installs. Add three lines to your agent loop. TypeScript/OpenClaw via HTTP sidecar. MIT licensed.

DSMC Starter Bundle

AGENT_MANIFEST v2.7 + MAGUS Philosophy v3.0 — paste into any frontier model and ask it to evaluate the architecture. Four models did. Same verdict each time.

DSMC Prompt Foundations

Four-layer prompting guide — Basic Prompts, Prompt Engineering, Context Engineering, Token Optimisation. Plain language. No code. Four ready-to-use templates included.

Agent Builders

Python implementations — production governance with persistence, drift detection, and dashboard

OpenClaw Agent Control Guide: API Edition v1.3.1

CDE classifier + confidence scoring + SQLite Revision Trail + structural and semantic drift detection + Gradio dashboard + Node.js/TypeScript sidecar — Anthropic, OpenAI, Gemini

Local LLM Agent Control Layer v1.3.1

Fully offline governance + confidence scoring + SQLite + ChromaDB + Gradio dashboard + Node.js/TypeScript sidecar — Ollama, LM Studio, llama.cpp. Zero cloud dependencies.

Mode Series

Workflow-specific governance — reconfigured DSMC for three session types

DSMC Mode Bundle

Three complete governance implementations: Generative Session Protocol (prevents premature convergence) + Epistemic Engineering Protocol (prevents epistemic flattening) + Brief Integrity Protocol (prevents brief erosion)

Architecture Series

MAGUS v3.0 — Agent/API pathway complete — enquiry via email

MAGUS v3.0 Architecture Series — Docs 2–5 (Agent/API)

Guardian (zero-trust execution governance + DEL), Stability Envelope, Agent Taxonomy A0–A5, GSTH, Cumulative Risk Score — ~38,600 words across four documents. Contact for pricing and delivery.

[email protected]

Want to try before you buy? The free Token Efficiency Suite on GitHub gives you working active state governance in five minutes — zero dependencies, MIT licensed. When you need cross-session persistence, drift detection, and a production dashboard, upgrade to the paid guides.

Not sure which guide? API Edition if you're using Anthropic, OpenAI, or Gemini. Local LLM Edition if you're running Ollama, LM Studio, or llama.cpp. Both include Node.js/TypeScript sidecars for OpenClaw — no Python in your stack required. Join the OpenClaw community.

Crypto payments / direct acquisition: Contact va@vahive.co — BTC, ETH, USDT accepted, files delivered on confirmed receipt.