Google Jules Swarm Suite

Jules Swarm Suite

Google Jules gives you cloud coding agents. But they freeze while you sleep.

Google Jules is one of the most capable autonomous coding systems in the world, running directly on your GitHub repositories in Google’s cloud. Out of the box, scaling it hits three bottlenecks:

The Human-in-the-Loop Freeze: Once Jules generates a plan (AWAITING_PLAN_APPROVAL) or needs architectural clarification (AWAITING_USER_FEEDBACK), your cloud agent pauses and waits. If you’re away from your desk, that worker sits idle for hours. The 15-Session Concurrency Cap: Google limits each account to 15 concurrent active sessions. The 100-Session Daily Wall (HTTP 429): Once an account hits 100 runs in a 24-hour rolling window, Google halts execution until the quota resets.

Jules Swarm Suite turns Google Jules into a self-healing, 24/7 autonomous coding factory. It federates multiple Google accounts into a pooled 500-session daily ring and maintains exactly 15 active cloud sessions non-stop. It auto-answers Jules’s technical questions in seconds via OpenRouter under strict enterprise rules, and merges verified PRs across all your GitHub mirrors.

How the Swarm Works

Point your agent at any GitHub repository. Jules Swarm Suite establishes a remote control plane over the Google Cloud REST API across six operational dimensions:

• Dimension: Weight, What it does
• Steady-state concurrency maintenance: 25%, Continuously monitors Google’s cloud API and replenishes vacant slots to keep all 15 cloud workers saturated 24/7.
• Autonomous feedback arbiter: 20%, Intercepts AWAITING_USER_FEEDBACK, extracts the agent’s technical question from the cloud activity log, synthesizes an actionable answer via OpenRouter under Ponytail Hacker rules (files ≤350 LOC, zero test diffs), and pushes :approvePlan in seconds.
• 5-account circular ring & 429 failover: 20%, Aggregates 500 sessions per day (100/account/day) across 5 federated worker accounts. When an account hits quota, an active drain barrier waits for running tasks to finish before rotating credentials.
• PR harvesting & 3-way merge: 15%, Fetches completed cloud branches from GitHub, runs dual-pass LLM safety audits (Reviewer + Verifier), executes 3-way git merges, and strips merge conflicts automatically.
• Multi-remote mirror sync: 10%, Pushes verified master and main branches to all 4 configured Git remotes (origin, mirror-arketek, mirror-jules, mirror-wellness).
• Diff-driven persona planning: 10%, Inspects your local git diff against remote master and maps modified files to specialized engineering personas (Sentinel for Auth, Cipher for Crypto, Gatekeeper for APIs, Optimizer for Performance).

12 Playbooks, One Package

• Command: Description
• /jules daemon: 24/7 continuous autonomous maintainer with a live web cockpit on port 8090.
• /jules sweep: 60-second single-pass reconciliation: audits cloud sessions, harvests patches, merges, and pushes.
• /jules clarify: Autonomous feedback arbiter. Unblocks all waiting cloud sessions in under 15 seconds.
• /jules plan: Analyzes repository git diffs and dispatches matched domain personas directly to Google Cloud.
• /jules reap: Extracts unidiff patches from completed cloud sessions with dual-pass safety auditing.
• /jules sync: Merges review branches --no-ff and synchronizes across all configured Git mirrors.
• /jules quota: Real-time 5-worker quota ledger report with remaining pool capacity and reset countdowns.
• /jules drain: Active drain barrier. Waits for in-flight tasks to hit 0 before rotating accounts, preventing dropped sessions.
• /jules audit: Pulls enterprise compliance playbooks from Google Workspace Docs via GWS CLI and dispatches audits.
• /jules prune: Zombie session killer. Prunes stuck cloud sessions older than 4 hours to reclaim concurrency slots.
• /jules secrets: Validates Bitwarden Secrets Manager (BWS) tokens and Google API keys.
• /jules compare: Weekly productivity report showing PR throughput, quota velocity, and token savings.

Built for Scaling Autonomous Engineering Interactive Cockpit UI (/jules web): Standalone FastAPI/Tailwind real-time dashboard on port 8090 tracking all 15 cloud slots, active branches, live logs, and quota gauges. Universal Zero-Config Portability: Zero hardcoded paths. Extracted into any directory on macOS, Linux, or Windows, it auto-detects your git repository root and manages state inside $PROJECT_ROOT/.jules/. White-Labeling Included: Configure custom agency names, remote mirror URLs, and telemetry webhooks by editing one file (white-label/brand_config.py). Who It Is For Software Agencies & Consultancies: Deliver client codebases and complex refactors 5x faster by keeping 15 cloud agents coding around the clock without manual supervision. Solo Developers & Startup Founders: Gain the output of a 15-person engineering team running background tasks, test writing, and bug fixes directly against your GitHub repos. High-Velocity Open-Source Maintainers: Automate dependency updates, linting fixes, documentation syncing, and PR reviews without getting rate-limited. How It Is Built

A single root SKILL.md acts as an intelligent router using progressive disclosure, keeping your prompt context under 2,500 tokens. It ships with:

6 Agent Persona Specifications (agents/) 5 Input/Output Templates (examples/) showing the agent what good code reviews and diffs look like 7 Deep Technical References (references/) 6 Paste-Ready JSON Schemas (schema/) 6 Modular Python Engines (scripts/), each strictly bounded to ≤350 lines of code FastAPI / Tailwind Web Cockpit (scripts/webapp/) 2 Printable & PDF Audit Templates (templates/)

Works with zero installs. Core cloud polling and orchestration run out of the box using standard Python 3.10+ and Node.js. Optional FastAPI/Uvicorn enables the real-time visual web cockpit.