ADE is a desktop IDE for macOS and Linux built around Claude Code, with Codex and local models running alongside it. Every project gets a methodology and a team of agents, and you see exactly what each one costs. It goes furthest on serious engineering work: nineteen roles that cover a whole company, a verification gate that decides when a story is actually done, and smart contracts from first test to reviewed deploy.
AI Agents
- Many agents at once: run Claude Code, Codex or any CLI agent in parallel tabs and split panes. The fleet view tracks every agent and sub-agent on a live timeline with its real cost.
- 47 agent profiles: Backend (including senior Go and Rust engineers), Frontend, Mobile, Data, DevOps, Testing, Security, Web3, Architects and General. Pick one and run it in this terminal or a new tab.
- One role, any CLI: a profile composes one role definition in
~/.ade/roles/, shared by every project; Claude Code and Ollama (deepseek and other local models) each receive it the way they accept one. Codex is shown unavailable rather than guessing a flag that fails silently.
- A whole company, not just a delivery team: nineteen roles end to end, from analyst and product manager through developer and QA to security engineer, SRE, release manager, support and solutions engineering. The picker lists every one unpaired, so the roles no profile covers are still a keystroke away.
- Project setup: every project is set up for agents once, automatically. See below.
- The agents are a repo you can share: every role definition lives in ade-setup, versioned and released, and ADE ships a pinned copy. Improve a role there and it reaches your projects when you choose to update them, not on ADE's release schedule.
- Prompt scratchpad: draft long prompts under every terminal with prompt tips and de-slop, send with one key, chain steps and reuse from history. Drafts survive crashes.
- Tokens & cost: spend at API prices, cache savings and context meters, plus one-click token savers (compact paste, context guard, context diet).
- Anti-slop: the ADE plugin for Claude Code (anti-slop skill, a slop check before Claude finishes,
/ade:deslop, Web3/Go/Rust sub-agents and architect skills), plus a slop check on your uncommitted changes.
- MCP library & Secrets vault: 15 curated MCP servers added to
.mcp.json in one click, with ${NAME} references to secrets kept in your OS keychain and never synced.
Project setup
Every project ADE opens (New project, Open project, or any git repo a terminal enters) is set up once:
- BMAD, on every project and actually used: the pinned BMAD-METHOD v6 by default, with its skills in
.claude/skills/, config in _bmad/ and tickets in _bmad-output/. v6 dropped BMAD's QA gate, so ADE keeps its own, one verdict per ticket in .ade/gates/. Eight roles name the skills that do their job instead of improvising, and v6's five persona skills are installed with the rest. v4 (.bmad-core/ with its tasks, checklists, templates and workflows, no personas) stays a per-project choice.
- The ADE methodology, "verified, not vibed": Plan → Approve → Shard → Build → Verify. A story is Done only when the agreed verification command passes, and no agent certifies its own work. The rules live in
.ade/rules.md, loaded from CLAUDE.md, alongside a context store, a decision log and a session journal.
- Eight core roles as Claude Code sub-agents in
.claude/agents/: product manager, architect, designer, scrum master, developer, QA, DevOps and adversarial reviewer.
- A knowledge store the agents write to:
.ade/knowledge/<role>.md, committed with your repo. A role records what it learned here (a flaky suite to serialise, a build step with a hidden prerequisite) and reads it back next run instead of rediscovering it. An update never touches it.
- Agents for your stack, detected from the project: Solidity and audit agents for Foundry or Hardhat, Anchor and Rust agents for Solana, a senior Go engineer for
go.mod, a senior Rust engineer for Cargo.toml.
Only missing files are written, and the toast that lists them has Undo. Add or remove agents any time in Integrations → Agents; turn automatic setup off in Settings → Terminal. See the project setup guide.
Blockchain Engineering Tooling
- Contracts panel: one key opens build, test, gas report, coverage, Slither and Aderyn analysis and a local chain for Foundry, Hardhat and Anchor projects.
- Workbench: run Foundry tests one at a time with gas, fuzz runs and counterexamples, then deploy and call contracts on Anvil with decoded events and reverts.
- Reviewed deploys: real-network deploy commands are typed into the terminal for you to review, never run automatically. ADE never handles, stores or syncs private keys.
- ABI viewer: functions, events and errors with their selectors, and Anchor IDLs with their accounts.
- Web3 agents: Solidity engineer (test-first, fuzz and invariant tests), smart contract auditor (reentrancy, access control, oracle and MEV, with proof-of-concept tests), gas optimizer, Solana/Anchor engineer, Web3 DevOps, and architects for DeFi, tokenomics, infrastructure and cross-chain.
Also in ADE
- Clickable files, live preview and an editor: Markdown with Mermaid, PDF, Word, images and HTML open beside the terminal; a source file opens in a Monaco editor tab that saves back to disk, with a live diagram pane for
.mmd files.
- Voice dictation: talk into the scratchpad instead of typing a long prompt.
- Auto-save and sync: sessions restore after a crash; settings, notes and Claude memory sync through your own private git repo.
- Precision, the default theme, with eight more presets (GitHub Dark, Dracula, Monokai Pro, Nord, Catppuccin Mocha, Solarized Dark, Tokyo Night, One Dark), per-colour overrides and a terminal palette that stays legible against the UI. The saved theme paints before the first frame, so a cold start never flashes.
Install
curl -fsSL https://ade.ardata.tech/install.sh | bash
Works on macOS (Apple Silicon and Intel) and Linux x86_64. It picks the right
build, checks it against the release's SHA256SUMS, and installs without sudo.
Pass --dry-run to see what it would do, or --help for the options.
Installers: download from the latest release: .dmg for macOS (Apple Silicon and Intel) and .deb or .AppImage for Linux.
ADE works with Claude Code, Codex and Ollama. See the user guide to get started.
Licence
ADE is free and open source under the MIT licence. Every feature, no account, no keys, no telemetry. Use it at work, fork it, ship your own build, no permission needed.
Issues and pull requests are welcome, on the app and on ade-setup, where the agent definitions live. CONTRIBUTING.md has the build steps, what CI checks, and a path in that needs no Rust: the nineteen roles are markdown.
Used on real work
ADE is the tool behind ardata.tech's projects. The agent fleet, the methodology and the contract tooling exist because client work needed them, not as a demo. It is also the tool used in ardata academy's AI training, the same setup in every lesson.
More on the website.
Support
For help, licensing and team purchases, get in touch via the website. The troubleshooting guide covers common issues.
© 2025–2026 Alvin Reyes · MIT
