Acolyte

English | 中文

Acolyte is a content analysis and evaluation system that detects bias, misleading content, and hidden intent in text. Submit content via Web UI, CLI, or API — analyzed by single or multiple LLMs.

Features

• Web UI, CLI, and API access
• Multi-LLM support: Anthropic Claude, OpenAI, Google Gemini
• Analysis modes: single LLM, multi-LLM parallel, multi-LLM with review/voting
• Scoring: bias index, misleading index, hidden intent index, credibility score
• History: task records, result visualization
• Modern web stack: React + TypeScript + Tailwind CSS v4 + shadcn UI

Quick Start

Prerequisites

• Python 3.10+
• uv
• bun

Backend

uv sync --extra dev   # install all dependencies
uv run acolyte        # start API server (http://localhost:8000)

Frontend

cd acolyte/web
bun install           # install dependencies
bun run dev           # start dev server (http://localhost:5173)

CLI

uv run acolyte analyze content.txt --mode=single
uv run acolyte history list --limit=5
uv run acolyte config add-llm -n "My LLM" -k "sk-..." -u "https://api.openai.com/v1" -m "gpt-4o"

Project Structure

acolyte/
├── acolyte/              # backend Python package
│   ├── core/             # LLM, DB, task processing
│   ├── api/              # FastAPI routes
│   ├── cli/              # CLI (Click)
│   └── web/              # React + Vite frontend
├── tests/                # Python tests (424 passed)
├── prompt/               # bias detection prompt templates
├── tools/                # utility scripts
├── pyproject.toml        # project config + dependencies
├── uv.lock               # pinned dependency versions
└── Makefile              # convenience commands

Configuration

Config file at ~/.config/acolyte/config.json (override with ACOLYTE_CONFIG_PATH):

{
  "database_url": "sqlite:///acolyte.db",
  "default_prompt_version": "",
  "llm_configs": [
    {
      "name": "Claude-Sonnet",
      "api_key": "your-api-key",
      "base_url": "https://api.anthropic.com/v1",
      "model_name": "claude-sonnet-4-20250514",
      "role": "normal",
      "is_default": true
    }
  ]
}

Development

# Backend
uv sync --extra dev     # install with dev deps
uv run pytest           # run tests (424 passed)
uv run ruff check .     # lint
uv run black .          # format

# Frontend
cd acolyte/web
bun run dev             # dev server
bun run build           # production build
bun run test            # run tests (26 passed)
bun run lint            # ESLint

# Convenience (from repo root)
make test               # backend tests
make web-test           # frontend tests
make lint               # backend lint
make clean              # remove build artifacts

Prompt Templates

Bias detection prompts in prompt/ follow the naming convention bias-detection-prompt_vX.Y[_modelname].md. After adding prompts, sync to database:

uv run acolyte config sync-prompts

License

MIT