# entroly — Context Guardrails to Catch Hallucinations

> Entroly is a local context compressor that traces claims back to real code to catch hallucinations and shrink prompts. pip install then entroly go.

## Install

Merge the JSON below into your `.mcp.json`:

## Quick Use

```bash
pip install entroly
cd /path/to/your/repo
entroly go
```

## Intro

Entroly is a local context compressor that traces claims back to real code to catch hallucinations and shrink prompts. pip install then entroly go.

**Best for:** Teams who want a local way to reduce hallucinations and token burn

**Works with:** Python 3.10+, pip, popular coding agents (Claude/Cursor/Codex/Copilot)

**Setup time:** 1-4 minutes

### Key facts (verified)

- GitHub: 356 stars · 62 forks · pushed 2026-05-13.
- License: Apache-2.0 · owner avatar + repo URL verified via GitHub API.
- README-verified entrypoint: `pip install entroly && cd /path/to/your/project`.

## Main

- Use it as a reality check for agent output: if the answer mentions a function, file, or API, Entroly helps trace it back to a real source location before you ship the change.

- Start with the default workflow (`entroly go`) to get a dashboard view, then iterate on budgets once you’ve seen savings on your real repo.

- Treat the README’s numbers as a hypothesis you can verify: it includes a local verification/benchmark flow you can run in your own environment.

### Source-backed notes

- README headline claims 70–95% token savings and a “30-second install”, and shows a pip-based install + `entroly go` flow.
- README includes a reproducible verification section that runs a claims-verification script against a repository.
- README includes latency and benchmark tables (e.g., token savings and retrieval results) to back its compression claims.

### FAQ

- **Does it require API keys?**: Not for the local compression/verification flows in README; any model calls depend on your chosen agent stack.
- **Is it only for Python repos?**: No — README claims it supports multiple file types and project layouts; validate on your repo.
- **How do I start safely?**: Run `entroly go` on a small repo first, then widen scope once you trust the outputs.

## Source & Thanks

> Source: https://github.com/juyterman1000/entroly
> License: Apache-2.0
> GitHub stars: 356 · forks: 62

---

<!-- ZH -->

## Quick Use

```bash
pip install entroly
cd /path/to/your/repo
entroly go
```

## Intro

Entroly 是本地上下文压缩器：把回答回溯到真实源码，拦住“凭空捏造”的 API/函数，并把提示词体积压到更小预算。README 宣称可省 70–95% token，pip 安装后进仓库跑 `entroly go`。

**Best for:** 想用本地方案降低幻觉与 token 消耗的团队/个人

**Works with:** Python 3.10+；pip；Claude/Cursor/Codex/Copilot 等常见编码工具

**Setup time:** 1-4 minutes

### Key facts (verified)

- GitHub：356 stars · 62 forks；最近更新 2026-05-13。
- 许可证：Apache-2.0；作者头像与仓库链接均已通过 GitHub API 复核。
- README 中核对过的入口命令：`pip install entroly && cd /path/to/your/project`。

## Main

- 把它当作 agent 输出的“事实校验层”：当回答提到某个函数/文件/API 时，先让它回溯到真实源码位置再动手改代码。

- 先用默认流程（`entroly go`）跑起来看面板，再在你自己的仓库里逐步调整预算与策略。

- 把 README 的数据当作可验证假设：它提供本地可复现的验证/基准脚本，你可以在自己的仓库复跑。

### Source-backed notes

- README 标题区写明可省 70–95% token，并给出 pip 安装 + `entroly go` 的快速启动方式。
- README 提供“可复现验证”章节：用脚本对仓库进行 claims verification。
- README 给出延迟与基准表（如 token savings、检索对比）来支撑其压缩效果主张。

### FAQ

- **需要 API key 吗？**：README 的本地压缩/验证流程不需要；是否调用模型取决于你接入的 agent/客户端。
- **只适用于 Python 仓库吗？**：不是。README 声称支持多文件类型/多项目布局；建议在你的仓库里验证。
- **怎么安全开始？**：先在小仓库跑 `entroly go` 熟悉输出，再逐步扩大到主仓库/单体项目。

## Source & Thanks

> Source: https://github.com/juyterman1000/entroly
> License: Apache-2.0
> GitHub stars: 356 · forks: 62


---
Source: https://tokrepo.com/en/workflows/entroly-context-guardrails-to-catch-hallucinations
Author: MCP Hub