Thirteen personas
From staff engineer to fractional CTO; bin/init writes the hub for one client.
Verified, not vibed · for Claude Code developers and architects
A team of agents. One gate for Done.
Your coding agents as BMAD v6 roles. They plan, a reviewer tries to break the plan, developers build in parallel git worktrees, and one verification command decides Done.
- — stars
- — forks
- — watching
Orchestration
A team of agents, one plan.
BMAD v6 decides who does what, one command decides Done, and the agent hub remembers it all.
Agent hub
A hub that remembers each client’s work.
One private repo per client holds its context, state and history. The client’s code stays in its own repos.
Focus terminals
bin/work <folder> opens that checkout with its named agent, hooks and model (--model, --agent-model).
State, memory, journal
STATE.md, memory/ and a shared journal; hooks show each terminal what the others changed.
One brain for every agent
A local SQLite brain (FTS5 plus Ollama vectors) that every agent searches before it asks; nothing leaves the Mac.
Auto sync, scanned first
bin/sync --auto pushes every 30 minutes and at session end, gitleaks-scanned, never forced.
Monitors and briefing
Read-only monitors every 30 minutes and a weekday morning briefing on Haiku.
Access and secrets
bin/remote for SSH, bin/secret for the Keychain, bin/doctor for what this Mac can reach.
BMAD v6 inside
BMAD v6 skills ship in the hub, under the ADE rules.
The full guide
Setup, the daily flow, the brain and troubleshooting.
Read the agent hub guide Or take the 15-minute training
Features
AI agents and blockchain tooling, in one window.
Orchestration, cost tracking and a contract toolchain that never touches private keys.
01
The terminal
Any CLI agent, side by side, with cost and context per session.
Every project, set up for agents
BMAD v6 or v4, the ADE methodology, eight core agents and agents for your stack.
On v6, CLAUDE.md tells the project’s agents to use the v6 skills. A CLAUDE.md you already have is never replaced.
Claude Code, first class
Sub-agents, a CLAUDE.md constitution, real spend from its transcripts, a plugin with 7 agents and 13 skills, and a tab dot when an agent needs you.
acme-api
~/code/acme-api
2 running
$2.75
- developer
- ↳ Explore
- ↳ Plan
- ↳ adversarial-reviewer
- qa
- ↳ Explore
Every agent on one fleet view
Every agent and sub-agent on one live timeline, with its real cost.
66 agents, one keystroke away
47 role-and-domain profiles and 19 standalone roles.
A scratchpad for prompts
Draft, chain and reuse prompts. Drafts survive crashes.
Cost and context per session
Spend, cache savings and context use, with one-click fixes.
02
A methodology that decides Done
An agent saying it is finished is not evidence. Every project names who checks the work.
ADE sets the rules; BMAD supplies the workflows.
BMAD in every project, wired into the roles
The pinned BMAD-METHOD v6 goes into every project: 33 skills in .claude/skills/, config in _bmad/, tickets in _bmad-output/. ADE keeps its own QA gate in .ade/gates/, and nothing you have is overwritten.
BMAD v4 is still a choice. Switching either way is one dialog that lists every path first, archives the old version in .ade/archive/ and puts everything back if a step fails.
Done is decided by a command
A story is Done when the agreed verification command passes on its commit, output shown.
No agent certifies its own work
The Developer writes the failing test first; QA judges and may not edit code or tests.
An adversarial reviewer that cannot patch
A read-only allowlist, Read, Grep, Glob, Bash: it judges, it never patches.
What they learn stays in your repo
A knowledge store per role and a shared context store, committed with the project.
03
How an agent is composed
A role fixes what an agent is accountable for; a domain narrows what it works on.
Role
19
What it owns, and what it must not touch.
Domain
44
One focus, like Backend API or Solidity. It never widens what the agent may do.
Agent
47
Named pairings: Developer × Backend API is the API Builder.
The role reaches the provider verbatim, with the domain appended as a Focus section, so its boundaries stay intact.
The team
Nineteen roles, and you approve the plan
Younothing ships without this
Delivery
Analystreads the problem
Product Managerowns the PRD
Designerowns the UX spec
Architectnames the verification command
Product Owneraccepts the work
Scrum Mastershards the plan
Developerwrites the test first
QAjudges, cannot edit
DevOpsowns the ops plan
Adversarial Reviewerread-only, defaults to block
Around it
Security Engineerthreat surface
Site Reliability Engineerrunbooks
Release Managerwhat ships, when
Engineering Managerload and sequencing
Support Engineerwhat users hit
Solutions Engineerwhat buyers need
Advisory
Brainstorming Architectshapes the idea
Technical Writerowns the docs
Advisorsecond opinion
Execution
Where each one acts
The agent that builds a story never signs it off.
-
Plan
PRD, architecture with the one verification command, UX spec and ops plan.
-
Approve
The adversarial reviewer attacks each artifact; then you approve.
-
Shard
The scrum master splits it into small, testable stories.
-
Build
Developers take one story each, test first, in parallel.
-
Verify
QA or you run the command. It passes, or it is not Done.
04
Agentic workflow and blockchain engineering
Architects for agents, RAG and LLMOps beside Foundry, Hardhat and Anchor. ADE never touches private keys.
One shortcut for the whole toolchain
Build, test, gas, coverage, Slither and a local chain.
A workbench for tests
Gas, fuzz runs and counterexamples per test; deploy and call on Anvil.
Reviewed deploys, readable ABIs
Deploys are typed for you to review, never run. ABIs with selectors.
Architects for agents and for chains
AI agent, RAG and LLMOps beside DeFi and cross-chain, plus an AI × Web3 architect.
MCP servers, from a list
GitHub, Playwright, Postgres, Context7 and more; keys go to the vault.
Anti-slop, on the lines just written
TODOs, debug leftovers, stubs and AI-sounding prose, flagged on the diff and in a Claude Code hook.
New
v0.22.0: the code viewer: click path:line in a terminal to open that file at that line. Release notes. Recently: sub-terminals (⌘[ ⌘]), a busy check before closing, the agent hub guide, BMAD v4/v6 switching, the stage board, Anti-slop, the MCP library and the Secrets vault.
Downloads
Get ADE
One command, or pick a build. The installer checks published checksums and needs no sudo.
curl -fsSL https://ade.ardata.tech/install.sh | bash
macOS and Linux. On Windows, see below.
- macOSApple Silicon .dmg
- macOSIntel .dmg
- LinuxDebian, Ubuntu .deb
- LinuxAppImage, any distro .AppImage
- WindowsNot built yet Track it —
Verify any download yourself against SHA256SUMS on the release.
Free and open source
Free, with no licence key, account or telemetry.
MIT licensed: use it at work, fork it, ship your own build.
- — stars
- — forks
- — watching
The whole app
One tier, and it is free.
Free forever
- Fleet view, tokens panel, scratchpad, editor and the contract workbench
- macOS (Apple Silicon and Intel) and Linux
- Keys and secrets stay in your OS keychain
The source
Read it, change it, lift what you want.
MIT licensed
- Tauri and Rust backend, React frontend
- Build and run your own version
- Issues and pull requests welcome
The agents
The nineteen role definitions live in their own repo.
Yours to fork
- Edit a role once, use it everywhere
- Versioned, so a project can pin what it uses
Used on real work
The tool its author ships with
ardata.tech
Visit ardata.tech
The tool behind ardata's projects
ardata builds its client work in ADE.
caiden.tech
Visit caiden.tech
The tool behind CAIDen
Voice, SMS and chat agents; Booked, NarraLearn and the Healthway front desk were built in ADE.
ardata academy
Explore the training
The tool behind ardata academy's AI training
Learn the workflow in the course, then use it on real projects.
Shipped with it
Products built using ADE
In production, built by the people who build ADE.
W3
The operating system for autonomous finance: build and control agent-powered financial workflows.
KONEKSI
Distributed-storage disaster recovery for files and databases, with automatic backup and verified restores.
DatArk Managed
Restores a real sample every month and publishes the verification report, rather than a backup-success log.
DeltaVault
Data disaster recovery that runs on-premises or in the cloud, complementing high availability.
CAIDen
AI agents for voice, SMS and chat, built and deployed per customer.
Healthway Front Desk
Answers patients on chat, Viber and the phone, and books their appointments.
Booked
One QR code across yard signs, reels, flyers and cards, turning every one of them into a booking.
NarraLearn
Turns a curriculum into a voice-narrated course with an AI tutor built in.
Crimson Security
Canadian cybersecurity firm: compliance, penetration testing and monitoring.
Wodan AI
Zero-trust AI.
Goldfish (GGBR)
A gold-backed token, over-collateralised beyond 100% by audited reserves.
Hoop Culture
Runs amateur basketball leagues, scoring games live by voice.
arena.stats
Leaderboards, rankings and match stats across every sport and registered athlete.
AR Code Scan
Scans code from GitHub, Replit or Lovable for security holes and production readiness.
Savvy
A real-estate assistant that answers on properties, neighbourhoods and market conditions.
IPFSR
A pluggable IPFS datastore that replaces Kubo flatfs and replicates to Sia renterd.
zipfs
IPFS implemented in Zig.
InSync
A personalised affirmations app for iPhone, read back in a clone of your own voice.
One Over Zero
Mobile, web and e-commerce builds alongside data-centre fit-out, blockchain and security work, from Makati.
FAQ
Common questions
Is it really free?
Yes. Download ADE; nothing is locked.
How does licensing work?
There is nothing to license. ADE is MIT licensed: no licence keys, accounts or seats.
Which agents does it support?
Any CLI agent; ADE goes deepest with Claude Code. Codex and Ollama run beside it but show no cost, because they publish none.
How many agents can run at once?
As many as your machine and rate limits allow. We test ten.
Does it store my keys?
API keys stay in your OS keychain, never synced. ADE never handles private keys.
Which platforms does it run on?
macOS (Apple Silicon, Intel) and Linux (.deb, AppImage), or the install command above. Windows is not built yet: issue #22.
Build your next product in ADE
Free and open source. Every feature, no account.