?? Palade

The AI-Powered Codebase Intelligence Engine

Palade isn't just a single bot. It's a highly orchestrated swarm of specialized AI agents that debate, arbitrate, and synthesize to find zero-day vulnerabilities and architectural flaws in your codebase.

?? Why Palade?

Traditional AI coding assistants send your entire file to a single language model. Palade uses a specialized multi-agent architecture to mimic a real software engineering team:

1. ??? Triage Agent: Scans your repository and figures out which files contain the most risk.
2. ??? Specialist Agents: We run multiple distinct agents concurrently over the same code:
   • Security, Architecture, Performance, Maintainability, Dead Code, and Test Intelligence.
3. ?? Arbitration Engine: If the Security agent and the Pragmatism agent disagree on a line of code, the Arbitration engine steps in to resolve the conflict.
4. ?? Synthesis Agent: Compiles the cross-cutting findings into a prioritized, actionable HTML and Markdown report.

?? Proven in Production
In rigorous benchmarking, the Palade Hybrid Swarm achieved 100% Precision and 100% Recall on known critical bugs. It even discovered and patched Zero-Day vulnerabilities in Axios and Zod that had bypassed thousands of human code reviews.

? Quick Start

Install Palade globally via npm:

npm install -g palade

Initialize Palade in your project to create the palade.config.ts:

palade init

Launch the interactive Terminal User Interface (TUI):

palade

?? Provider Support (100% Free Tier Available)

Palade runs entirely on your local machine and communicates directly with your LLM provider of choice. We support hybrid routing, meaning you can assign heavy tasks to frontier models and light tasks to hyper-fast local models.

| Provider | Environment Variable | Notes | | :--- | :--- | :--- | | OpenRouter | OPENROUTER_API_KEY | Best for heavy agents (e.g., tencent/hy3:free or claude-3.5-sonnet). | | OpenCode-Zen | OPENCODE_ZEN_API_KEY | Default. Excellent for syntax and fast checks (e.g., mimo-v2.5-free). | | Ollama (Local) | OLLAMA_MODEL | 100% private, offline inference. | | Groq / Cerebras | GROQ_API_KEY | LPU-powered execution for analyzing massive monorepos in seconds. | | Nvidia | NVIDIA_API_KEY | Access to massive parameter models like Nemotron. |

??? CLI Commands

You can run Palade directly from your terminal or CI/CD pipeline:

palade review

Run a full swarm review over your project.

palade review
palade review --file src/core/Auth.ts  # Target a specific file
palade review --target backend         # Review a pre-configured target area

Outputs a stunning interactive HTML report and a Markdown summary.

palade diff

Review only the files that have changed since your main branch or last commit. Perfect for CI pipelines.

palade diff --base main

palade watch

Run the review daemon in the background. It watches your files for changes, debounces them, and automatically runs a review.

palade watch

palade score

Check your codebase health score and view your score trajectory over time.

palade score

?? Advanced Configuration

The Hybrid Swarm

You can configure exactly which provider handles which agent in your palade.config.ts. For example, a $0 budget setup:

export default {
  swarm: {
    primary: "opencode-zen",
    agentProviders: {
      security: "openrouter",
      architecture: "openrouter",
    },
  },
  providers: {
    openrouter: { model: "nvidia/nemotron-3-ultra-550b-a55b:free" },
    "opencode-zen": { model: "mimo-v2.5-free" }
  }
}

Economy Mode

By default, Palade runs in Economy Mode, meaning all specialist lenses are evaluated in a single API call per chunk to drastically reduce API costs. If you want maximum prompt richness per domain, disable it in your config:

  swarm: {
    economyMode: false,
  }