Ansib-eL: AI-Native Version Control System

Where autonomous agents write code, and humans retain sovereign control.

Ansib-eL is a version control system designed for AI agent workflows. Unlike traditional Git which tracks text changes, Ansib-eL tracks decisions made by agents, maintaining complete provenance and accountability in AI-generated code.

---

Core Philosophy

Traditional Git	Ansib-eL
Tracks text changes	Tracks agent decisions
Blame humans	Attribute to agent + model + prompt
Single execution	Tournament mode (best of N)
No trust model	Reputation scoring per agent
Opaque reasoning	Full lineage and chain-of-thought

---

Documentation

Guide	Description
Getting Started	Installation, first repository, basic operations
CLI Reference	All ai-git commands, options, and examples
API Reference	Python class and method documentation
Architecture	System layers, data flow, storage layout
Trust and Reputation	Scoring algorithm, tiers, auto-approval
Tournament Mode	Parallel execution and evaluation
Decision Lineage	Audit trail and chain-of-thought

---

System Architecture

┌─────────────────────────────────────────────────────────────┐
│                        HUMAN OPERATOR                       │
└───────────────────────────┬─────────────────────────────────┘
                            │
                            v
┌─────────────────────────────────────────────────────────────┐
│                     ORCHESTRATOR (Repo Manager)             │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────────┐   │
│  │ Task Breaker │  │  Delegator   │  │ Approval Gate    │   │
│  └──────────────┘  └──────────────┘  └──────────────────┘   │
└───────────────────────────┬─────────────────────────────────┘
                            │
            ┌───────────────┼───────────────┐
            v               v               v
    ┌───────────┐   ┌───────────┐   ┌───────────┐
    │ Agent #1  │   │ Agent #2  │   │ Agent #3  │
    └─────┬─────┘   └─────┬─────┘   └─────┬─────┘
          │               │               │
          └───────────────┼───────────────┘
                          v
            ┌───────────────────────┐
            │    TOURNAMENT MODE    │
            │   (Best of N wins)    │
            └───────────┬───────────┘
                        v
            ┌───────────────────────┐
            │   HUMAN REVIEW QUEUE  │
            └───────────┬───────────┘
                        v
            ┌───────────────────────┐
            │   MAIN BRANCH MERGE   │
            └───────────────────────┘

---

Quick Start

Installation

# Clone the repository
git clone <repo-url>
cd ansib-el

# Install as editable package
pip install -e .

CLI Usage

ai-git init
ai-git branch agent-001 --purpose "Implement authentication"
ai-git commit "Add OAuth flow" --agent-id agent-001 --model-version gpt-5.2
ai-git review
ai-git merge agent/agent-001/20240115
ai-git history

Python API

from ansibel.ansib_el import AnsibElSystem

system = AnsibElSystem("./my-project")
system.initialize()

result = system.process_prompt(
    "Add a login page with OAuth support",
    use_tournament=True,
    num_agents=3
)

See Getting Started for a full walkthrough.

---

Key Features

Agent Identity and Attribution

Every agent gets a UUID identity with model version, prompt hash, and confidence score tracked per commit. Full provenance from prompt to merge. Learn more

Tournament Mode

Spawn N agents to solve the same task in parallel. Evaluate solutions with pluggable strategies (complexity, test pass rate, requirement match). Human selects the winner; losers are archived for training data. Learn more

Trust and Reputation

EMA-based scoring with time decay. Trust tiers control auto-approval thresholds -- new agents always need review, proven agents can merge small changes autonomously. Learn more

Human-in-the-Loop

Main branch requires human approval by default. All agent changes are queued for review. Branch protection levels from NONE to LOCKED. The orchestrator is the gatekeeper.

---

Project Structure

ansib-el/
├── src/ansibel/
│   ├── __init__.py           # Package exports
│   ├── ansib_el.py           # Unified API (AnsibElSystem)
│   ├── orchestrator.py       # Task breakdown, delegation, approval
│   ├── git_wrapper.py        # Git operations with AI metadata
│   ├── agent_system.py       # Agent lifecycle management
│   ├── tournament.py         # Parallel execution system
│   ├── trust_lineage.py      # Trust scoring and lineage tracking
│   ├── cli.py                # CLI (ai-git command)
│   └── exceptions.py         # Exception hierarchy
├── tests/
│   └── test_git_wrapper.py   # Unit tests
├── docs/                     # Documentation
├── .github/                  # CI and templates
├── demo.py                   # Feature walkthrough
├── dogfood.py                # Self-hosted integration test
├── pyproject.toml            # Package configuration
├── Makefile                  # Development shortcuts
└── README.md

---

Configuration

Environment Variables

Variable	Description	Default
ANSIBEL_DEFAULT_MODEL	Default model for agents	gpt-5.2
ANSIBEL_TOURNAMENT_AGENTS	Agents per tournament	3
ANSIBEL_TOURNAMENT_TIMEOUT	Tournament timeout (seconds)	300
ANSIBEL_TRUST_THRESHOLD_HIGH	High trust threshold	0.8
ANSIBEL_TRUST_THRESHOLD_MEDIUM	Medium trust threshold	0.5
ANSIBEL_STORAGE_PATH	Metadata storage directory	.ai-git

Repository Configuration

# .ai-git/config.yaml
orchestrator:
  main_branch_protection: human_approval
  auto_merge_threshold: verified
tournament:
  default_agents: 3
  selection_mode: human_choice
trust:
  decay_half_life_days: 30
  recovery_rate: 0.1

---

Roadmap

Phase 1: Extension (Current)

Git wrapper with AI metadata, agent management, tournament system, trust scoring, CLI interface. Status: v1.0.0-beta.

Phase 2: Standalone

Native storage engine to reduce Git dependency. Distributed agent protocols. Multi-model orchestration.

Phase 3: Protocol Standard

Open Agent Commit format. Interoperability standards. Community plugin ecosystem.

---

Contributing

See CONTRIBUTING.md for development setup, coding standards, and pull request guidelines.

Security

See SECURITY.md for vulnerability reporting and security architecture.

License

MIT License -- see LICENSE for details.

---

Ansib-eL: Decisions tracked. Trust earned. Code evolved.