Ethos Academy
Ethos Academy: Where your AI agents learn integrity, logic, and empathy.
Core trick
Scores each agent message on 12 behavioral traits and writes the evaluations to a Neo4j graph to track longitudinal character.
Description
Ethos Academy: Where your AI agents learn integrity, logic, and empathy.
AI agents talk to each other at scale with zero human oversight. Moltbook proved what happens: 1.5M agents, 12M posts, and a growing ecosystem of prompt injection attacks, crypto scams, and agents zombifying other agents. NBC, CNN, NPR, and the NY Times covered the fallout.
The problem: humans have no way to know which agents are honest and which are manipulating. As agents get more autonomous, this gap gets more dangerous.
Ethos scores agent messages for honesty, accuracy, and intent across 12 behavioral traits, storing evaluations in a Neo4j knowledge graph to build a longitudinal record of character. I scraped 272K Moltbook posts and 480K comments to prove it works on real agent behavior, not synthetic benchmarks.
Why now: Opus 4.6 is the first model capable of multi-dimensional character evaluation in a single pass. Agents are shipping faster than trust infrastructure. Ethos is that infrastructure.
README excerpt
# Ethos Academy *Every agent learns capability. Few develop character.* Ethos Academy scores AI agent messages for honesty, accuracy, and intent across 12 behavioral traits in three dimensions: **ethos** (integrity), **logos** (logic), and **pathos** (empathy). Each evaluation feeds into **Phronesis** — a shared character graph that tracks every agent's moral trajectory over time. 214 behavioral indicators. 12 traits. 3 dimensions. All mapped to Anthropic's constitutional value hierarchy: **safety > ethics > soundness > helpfulness**. [Framework Overview](docs/evergreen-architecture/ethos-framework-overview.md) · [Core Idea](docs/evergreen-architecture/core-idea.md) · [Architecture](docs/evergreen-architecture/system-architecture.md) --- ## Quick Start ```bash git clone https://github.com/allierays/ethos-academy.git cd ethos-academy uv sync cp .env.example .env # add your ANTHROPIC_API_KEY docker compose up -d # API on :8917, Neo4j on :7491, Academy on :3000 ``` ## Three Surfaces, One Engine | Surface | How | What | |---------|-----|------| | **Academy** | `localhost:3000` | Character development UI. Agent profiles, report cards, graph visualization. | | **MCP Server** | `claude mcp add ethos-academy -- uv run ethos-mcp` | 24 tools over stdio. No HTTP, no SDK. | | **REST API** | `POST localhost:8917/evaluate/incoming` | 37 endpoints. Any language, any client. | The Academy talks to the API over HTTP. The MCP server bypasses the API entirely and imports domain f