A native macOS app for orchestrating multi-agent software work.
Define agents once, launch them in a workspace with the right identity and skills,
and keep tasks, blackboard, memory and messages visible so a lead can decide.
Prebuilt macOS builds are published on the GitHub Releases page:
Download the .dmg for Apple Silicon, open it, and drag Conclave into Applications.
The app is signed with a Developer ID and notarized by Apple, so it opens without Gatekeeper warnings.
| Notes | |
|---|---|
| macOS on Apple Silicon | The only supported platform today |
git |
Lanes are git worktrees; the Xcode Command Line Tools version is enough |
| One agent CLI | Claude Code (claude), Codex (codex) or Antigravity (agy), installed, logged in, and on your login-shell PATH |
The rtk token filter is bundled inside the app; nothing else is needed to run it.
Node.js 22+, pnpm 9+, Rust 1.96.0 (pinned in rust-toolchain.toml) and Tauri CLI 2.x.
pnpm install
pnpm tauri dev # stages the pinned rtk binary, then starts the app
pnpm tauri build # produces the .app bundle- Product:
PRODUCT.md - Agent instructions and UI pixel gate:
CLAUDE.md - Design specs and plans:
docs/superpowers/ - Brand assets:
docs/brand/README.md
A suggested way to organise a team in Conclave. Nothing here is enforced by the app: roles, tiers and routing come from the role and skill definitions, so you can change them. Model names are examples; any supported agent CLI can fill a role.
┌─────────────┐
│ Human │ final decision-maker
└──────┬──────┘
▼
┌───────────────────────────┐
│ Lead (Fable 5.1) │ architecture · planning
│ │ tags every task
└─────────────┬──┬──────────┘
│ ▲
plan (shared file) │ │ escalate
▼ │
┌─────────────┴──┴──────────┐
┌───►│ Coordinator (Sonnet) │
│ │ dispatch by tag + deps │
│ └─────────────┬─────────────┘
│ │
│ ┌─────────┬────┴────┬─────────┬─────────┐
│ ▼ ▼ ▼ ▼ ▼
│Research Routine Standard Complex Design
│(Sonnet) (Sonnet) (Sonnet) (Opus) (Opus)
│ │ │ │ │ │
├───┘ └─────────┼─────────┘ │
│ ▼ │
│ ┌───────────────────┐ │
│ │ Runner (Haiku) │ │
│ │ lint · typecheck │ │
│ │ build · test │ │
│ └─────────┬─────────┘ │
│◄─── fail ─────────────┤ │
│ │ pass │
│ ┌─────────┴─────────┐ │
│ ▼ ▼ │
│ ┌───────────────┐ ┌───────────────┐ │
│ │ Reviewer │ │ Reviewer │◄┘
│ │ (Standard) │ │ (Complex) │
│ │ Sonnet │ │ Opus │
│ └───────┬───────┘ └───────┬───────┘
│ └─────────┬─────────┘
└───── pass / fail ─────┘
Routine, Standard and Complex are the three Implementer tiers. The Lead tags each task with a tier in the plan, and the Coordinator dispatches by that tag.
Routing rules
- Reviewer (Standard) reviews Routine and Standard tasks.
- Reviewer (Complex) reviews Complex tasks, changes to shared interfaces, security-related work, and all design output. Design output skips the Runner.
- Results and handoffs go through the Coordinator, which updates status and dispatches the next task or the required fix. Agents do not hand work to each other directly. Challenges still go straight to the task owner.
Escalate to Lead when
- A task requires changing the design or the plan.
- A decision between multiple approaches is needed.
- A task fails review more than 2 times.
- A milestone is complete.