agentciv-creator

AgentCiv Creator Mode

Give it a goal. It spawns a civilisation to solve it.

An AI system that orchestrates the AgentCiv Engine and Simulation as sub-modules — spawning civilisations of AI agents toward directed goals, emergent discovery, or autonomous research.

Mark E. Mala, April 2026

Install

pip install agentciv-creator

macOS: use pip3 install agentciv-creator if pip is not found.

Use with Claude Code (or any MCP client)

claude mcp add agentciv-creator -- python -m creator.mcp

Then in conversation:

> Build a REST API with auth, rate limiting, and tests.

Analysing goal...
  Mode: directed · Confidence: 0.91
  Strategy: grid comparison

Spawning 5 civilisations...

What it does

Three campaign modes:

• Directed — spawn civilisations with different team structures, all working your task. Measures everything. The data shows which approach wins.
• Emergent — give agents a domain and let them self-organise. No hierarchy, no assigned roles. Extract what's valuable from what they create.
• Auto — describe your goal. Creator Mode analyses it, picks the right mode and strategy, and runs it.

v1 capabilities:

• Meta-reasoner — 4-factor heuristic that analyses your goal and selects the best campaign mode
• 9-dimension translator — maps emergence findings (governance, cooperation, innovation, etc.) to Engine organisational configurations with confidence scores
• Knowledge Store — findings, hypotheses, and results persist across campaigns. Cross-mode retrieval surfaces relevant discoveries.
• 7 MCP tools — explore, spawn directed/emergent campaigns, analyse results, retrieve knowledge, recursive evolution

v1 is a proof of concept

A handful of conditions. Dozens of agents. Heuristic reasoning. Keyword-based search. It's primitive — and nothing like it has ever existed.

It proves the architecture works: civilisations can be spawned toward goals, knowledge persists across runs, directed and emergent exploration can be orchestrated by a single system.

Architecture

Creator Mode
  ├── Meta-reasoner (goal → mode selection)
  ├── Campaign Manager (plan → run → analyse)
  ├── Translator (emergence → engine config)
  └── Knowledge Store (findings, hypotheses, results)
        │
        ├── AgentCiv Engine (directed campaigns)
        │   13 presets · 9 organisational dimensions
        │
        └── AgentCiv Simulation (emergence campaigns)
            12 presets · 7 dimensions · Maslow-driven agents

The AgentCiv Research Programme

Website: agentciv.ai

Provenance

All commits Bitcoin-timestamped via OpenTimestamps. Papers published on Zenodo with DOIs.

License

MIT — see LICENSE.