Adaptive Multi-Agent (AMA)

What Adaptive Multi-Agent (AMA) does

The README is written in Chinese. AMA scores complexity from 1 to 10 using keyword features, implicit signals and an optional LLM refinement for ambiguous scores, and can ask clarifying questions first. It then maps the score to one of five modes: Generator-Verifier (1 to 3), Orchestrator-Subagent (4 to 6), Agent Teams (7 to 8), Message Bus (9) and Shared State (10). Mode choice mixes Thompson Sampling with Bayesian smoothing and a rule-based fallback, and results are saved in SQLite.

On failure it can switch modes, upgrading for verification or internal errors and downgrading for context overflow or timeouts, with a 30 second cooldown. Each mode has its own circuit breaker, which opens after five consecutive failures and recovers after 60 seconds. It exposes seven tools (ama_execute, ama_assess, ama_switch_mode, ama_stats, ama_cancel, ama_clarify and ama_diagnose) and exchanges task weights with a separate Model Router project.

Key features

• Five collaboration modes from Generator-Verifier to Shared State
• Complexity scoring from 1 to 10 with optional LLM refinement
• Thompson Sampling mode selection with SQLite-backed history
• Automatic mode switching on failure with a cooldown
• Per-mode circuit breakers
• Seven ama_* tools and session hooks

When to use it

• Let Hermes choose between a simple generate-and-verify loop and a multi-agent team based on the task
• Assess how complex a task is without running it
• Track which collaboration modes succeed most often over time

Who it is for: Hermes Agent developers who delegate tasks to subagents and want automatic selection of the collaboration pattern.

How it fits with Hermes Agent

Built as a Hermes plugin that uses delegate_task, plugin hooks and ctx.dispatch_tool. The README links to the author's own hermes repository and asks for Hermes 2.0.0 or newer, not to NousResearch/hermes-agent.

How to install Adaptive Multi-Agent (AMA)

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

git clone https://github.com/weksbwrx62862/adaptive-multi-agent.git
cd adaptive-multi-agent
pip install -e .

Requirements: Python 3.10+ and Hermes Agent 2.0.0 or newer

Note: The README ties the plugin to the author's own Hermes repository and Hermes 2.0.0 or newer, and the repository has no license file.

FAQ