<!-- SEO / discovery metadata. GitHub renders this block invisibly; search engines, raw-file crawlers, and AI agents that read README.md pick it up. Repo TOPICS (set in repo settings) are the strongest GitHub-search signal - mirror these keywords there too. LucidAgentIDE - a fail-closed security, provenance, and memory layer around oh-my-pi (omp). Topics: ai-agent-security, prompt-injection-defense, llm-security, agent-observability, provenance, duckdb, kv-cache, prompt-caching, asksage, government-ai, cui, fips, personalization-knowledge-graph, ai-code-attribution, cost-showback, chatgpt-import, monaco, electron, bun, typescript, oh-my-pi, omp, agentic-coding, secure-ai-coding-assistant. --> <meta name="description" content="LucidAgentIDE - a fail-closed security, provenance, and memory layer around oh-my-pi (omp): prompt-injection defense, trust labeling, provenance-backed memory, sovereignty-aware model governance, AI-authorship attribution, one-command ChatGPT/Claude/Gemini migration, cross-model cost showback, and a read-write IDE where even Save is scanned." /> <meta name="keywords" content="AI agent security, prompt injection defense, fail-closed gate, Unicode scanner, LLM provenance, agent observability, DuckDB telemetry, KV-cache prompt optimization, prompt caching, AskSage government AI gateway, CUI, FIPS, personalization knowledge graph, AI code attribution, AI-authored lines of code, cross-model cost showback, ChatGPT import, Claude, Gemini, Monaco editor, Electron, Bun, TypeScript, oh-my-pi, omp, agentic coding, secure AI coding assistant, agent IDE" /> <meta name="robots" content="index, follow, max-image-preview:large" /> <meta name="googlebot" content="index, follow" /> <meta name="author" content="Nick Chadwick (TechLead187)" /> <div align="center"> <img src=".github/assets/techlead187-avatar.png" alt="TechLead187 - Nick Chadwick, creator of LucidAgentIDE" width="160" style="border-radius: 50%;" /> <br/> <img src=".github/assets/banner.svg" alt="LucidAgentIDE - a fail-closed security, provenance and memory harness around oh-my-pi" width="100%" /> <br/>

<a href="https://github.com/mlcyclops/lucidagentide/actions/workflows/ci.yml"><img src="https://img.shields.io/github/actions/workflow/status/mlcyclops/lucidagentide/ci.yml?branch=master&label=CI&logo=github&logoColor=white&style=flat-square" alt="CI" /></a> <a href="https://github.com/mlcyclops/lucidagentide/actions/workflows/codeql.yml"><img src="https://img.shields.io/github/actions/workflow/status/mlcyclops/lucidagentide/codeql.yml?branch=master&label=CodeQL&logo=github&logoColor=white&style=flat-square" alt="CodeQL SAST" /></a> <a href="https://github.com/mlcyclops/lucidagentide/actions/workflows/build-desktop.yml"><img src="https://img.shields.io/github/actions/workflow/status/mlcyclops/lucidagentide/build-desktop.yml?label=Windows%20Build&logo=windows&logoColor=white&style=flat-square" alt="Windows Build" /></a> <a href="https://github.com/mlcyclops/lucidagentide/actions/workflows/build-desktop.yml"><img src="https://img.shields.io/github/actions/workflow/status/mlcyclops/lucidagentide/build-desktop.yml?label=macOS%20Build&logo=apple&logoColor=white&style=flat-square" alt="macOS Build" /></a> <a href="https://github.com/mlcyclops/lucidagentide/actions/workflows/build-desktop.yml"><img src="https://img.shields.io/github/actions/workflow/status/mlcyclops/lucidagentide/build-desktop.yml?label=Linux%20Build&logo=linux&logoColor=white&style=flat-square" alt="Linux Build" /></a> <img src="https://img.shields.io/badge/tests-1803%20harness%20%2B%204280%20desktop%20%2B%2057%20sidecar-46d27e?style=flat-square" alt="tests" /> <img src="https://img.shields.io/badge/gate-fail--closed-e07bf0?style=flat-square" alt="fail-closed gate" />

<br/>

<a href="https://github.com/mlcyclops/lucidagentide/releases/latest/download/LucidAgent-Setup.exe"><img src="https://img.shields.io/badge/Download-Windows%20Installer-2ea44f?style=for-the-badge&logo=windows&logoColor=white" alt="Download Windows installer (latest release)" /></a> <a href="https://github.com/mlcyclops/lucidagentide/releases/latest/download/LucidAgent-mac-arm64.pkg"><img src="https://img.shields.io/badge/Download-macOS%20Apple%20Silicon%20(.pkg)-000000?style=for-the-badge&logo=apple&logoColor=white" alt="Download macOS Apple Silicon .pkg installer (latest release)" /></a> <a href="https://github.com/mlcyclops/lucidagentide/releases/latest/download/LucidAgent-mac-x64.pkg"><img src="https://img.shields.io/badge/macOS-Intel%20(.pkg)-555555?style=for-the-badge&logo=apple&logoColor=white" alt="Download macOS Intel .pkg installer (latest release)" /></a> <a href="https://github.com/mlcyclops/lucidagentide/releases/latest/download/LucidAgent-x86_64.AppImage"><img src="https://img.shields.io/badge/Download-Linux%20AppImage-f0a500?style=for-the-badge&logo=linux&logoColor=white" alt="Download Linux AppImage (latest release)" /></a> <a href="https://github.com/mlcyclops/lucidagentide/releases/latest"><img src="https://img.shields.io/github/v/release/mlcyclops/lucidagentide?label=latest&style=for-the-badge&color=c64bd6&sort=semver" alt="Latest release version" /></a>

_{⬆ Always the most recent successful release - links auto-update each version (no release yet? they appear after the first tagged build).}

<br/> <img src="https://img.shields.io/badge/Bun-%E2%89%A51.3-fbf0df?style=flat-square&logo=bun&logoColor=black" alt="Bun" /> <img src="https://img.shields.io/badge/TypeScript-3178c6?style=flat-square&logo=typescript&logoColor=white" alt="TypeScript" /> <img src="https://img.shields.io/badge/Electron-47848F?style=flat-square&logo=electron&logoColor=white" alt="Electron" /> <img src="https://img.shields.io/badge/DuckDB-FFF000?style=flat-square&logo=duckdb&logoColor=black" alt="DuckDB" /> <br/>

A security · provenance · memory layer built around <a href="https://omp.sh">oh-my-pi</a> - not a fork. A fail-closed prompt-injection gate, provenance-backed memory, sovereignty-aware model governance, AI-authorship attribution, one-command migration from ChatGPT, and a read-write IDE where even Save is scanned - wrapped in a polished desktop app, added entirely through omp's hooks, custom tools, and SDK.

_{🔒 <b>What it does is open; how the hard parts work is not.</b> The deepest trust, provenance, and personalization internals are proprietary and intentionally undocumented here - this README describes the <i>capabilities and guarantees</i>, not the mechanisms behind them.}

<a href="#-who-its-for"><b>Who it's for</b></a> · <a href="#-quick-start"><b>Quick start</b></a> · <a href="#-security-model"><b>Security</b></a> · <a href="#-token-cost-savings--showback"><b>Cost Savings</b></a> · <a href="#-knowledge--rag"><b>Knowledge / RAG</b></a> · <a href="#-contributing"><b>Contributing</b></a> · <a href="#-roadmap"><b>Roadmap</b></a> · <a href="DECISIONS.md"><b>Decisions (ADRs)</b></a>

<br/> <!-- v2.3 RELEASE BANNER --> <table align="center" width="100%"> <tr> <td align="center">

🚀 LUCID v2.3.1 is here

<p align="center"><b>🛡️ a Windows sandbox that chat actually works inside</b> · <b>🧭 Jev typed judgment</b> · <b>🖱 a browser agent that reads before it clicks</b> · <b>🪐 Claude Opus 5.5 and Sonnet 5.5, GPT-6 Astra / Sol / Luna, Grok 4.7</b> · <b>📶 network problems reported as network problems</b> · <b>🌳 fleet lanes in their own worktree</b> · <b>🖥 <code>lucid hub</code> in your terminal</b> · <b>🕹 new in 2.3.1: four arcade games</b></p>

<a href="https://github.com/mlcyclops/lucidagentide/releases/tag/v2.3.1"><img src="https://img.shields.io/badge/Download-v2.3.1-c64bd6?style=for-the-badge&logo=github&logoColor=white" alt="Download LUCID v2.3.1" /></a> <a href="https://github.com/mlcyclops/lucidagentide/issues/new?template=bug_report.yml"><img src="https://img.shields.io/badge/Found%20a%20bug%3F-Open%20an%20issue-d73a49?style=for-the-badge&logo=github&logoColor=white" alt="Found a bug? Open an issue" /></a> <a href="#how-to-pull-your-logs"><img src="https://img.shields.io/badge/How%20to-pull%20your%20logs-555555?style=for-the-badge" alt="How to pull your logs" /></a>

<p align="center">_{v2.3.0 promoted fifteen public betas to stable; v2.3.1 adds four new games to Preview &gt; Games. The download buttons above, Homebrew and the in-app updater all move to 2.3.1. <a href="#-lucid-v230-is-here"><b>What's in it, how to update, and how to report a bug with screenshots and logs</b></a> · <a href="#-whats-new-in-v231"><b>What's new in 2.3.1</b></a>.}</p> <p align="center">_{<b>Next up: v2.4.0, as public betas first.</b> Like 2.3.0, the next release ships as GitHub prereleases (<code>v2.4.0-beta.N</code>) before anything reaches stable installs: the download buttons above, the in-app updater and the Homebrew cask stay on 2.3.1 until 2.4.0 is promoted. <a href="#-v240-ships-as-betas-first"><b>What the betas carry and how to track them</b></a>.}</p> </td> </tr> </table> <br/> <!-- HEADLINE BANNER - the current live model line-up --> <table align="center" width="100%"> <tr> <td align="center">

<img src=".github/assets/icons/announce-animated.svg" width="30" align="top" alt="" /> The newest frontier models are live in LUCID <img src=".github/assets/icons/announce-animated.svg" width="30" align="top" alt="" />

<b>Claude Opus 5.5 · Sonnet 5.5 · Opus 5</b>, <b>Claude Fable 5.1</b> and <b>Mythos 5.1</b>, <b>GPT-6 Astra · Sol · Luna</b>, <b>Grok 4.7</b>, and <b>Gemini 3.1 Pro · 3.8 Flash</b> - all in the picker today. And <b>Jev</b>, the <b>TypeSafe</b> typed-judgment engine, sits behind them as the judgment backend (beta).

<p align="center"><b>Connect the account you already pay for</b> (OAuth subscription or an API key) and pick the model from the list - that's it. Each one carries a <b>cost + intelligence card</b> and a clear <b>U.S.-government data-privacy notice</b>, so you always know what a turn costs and where your chat history stands.</p> <p align="center">_{<code>claude-opus-5-5</code> · <code>claude-sonnet-5-5</code> · <code>claude-opus-5</code> · <code>claude-fable-5-1</code> · <code>claude-mythos-5-1</code> · <code>gpt-6-astra</code> · <code>gpt-6-sol</code> · <code>gpt-6-luna</code> · <code>grok-4.7</code> · <code>gemini-3.1-pro</code> · <code>gemini-3.8-flash</code> · plus every other model the runtime exposes, including the AskSage gov gateway (<code>gpt-5.6-luna · sol · terra</code>, <code>google-gemini-3.1-pro-com</code>) and your own local endpoints.}</p> <p align="center">_{<b>Context windows are declared per model</b>, so Opus 5.5, Sonnet 5.5 and Opus 5, Fable/Mythos 5.1 and GPT-6 Astra, Sol and Luna are all billed and metered against their real <b>1M</b> window instead of inheriting a 256K assumption - and vendor-superseded ids are <b>removed</b> from the catalog rather than left to rot in the picker.}</p> <p align="center">_{<b>Jev / TypeSafe support</b> (<a href="#-whats-new-in-v230">new in v2.3.0</a>): paste a TypeSafe key under <b>Settings &gt; Judgment</b> and every typed judgment - classifications, yes/no checks, scores, the agent's <code>judge()</code> calls and the <code>browser_run</code> browser policy - is answered by Jev as a typed result with probabilities and a confidence, traced under the reply. No key, or AskSage lockdown, and the same calls fall back to your chat model.}</p> </td> </tr> </table> <br/> <h3 align="center">🌐 Also new - your agent is online out of the box, and every tool call works</h3> <p align="center"><b>“Allow all websites + local LAN” is pre-checked</b>, so a fresh agent can browse and search the web immediately - while the curated, trust-scoped whitelist is one toggle away whenever you want to lock it down (it still asks before a public IP or a foreign-country site).</p> </div>

---

🧪 v2.4.0 ships as betas first

**N