Jules is a suite of autonomous AI code review and engineering agents powered by LocalHarness. Jules agents autonomously inspect your target codebase, identify targeted domain-specific improvements, implement surgical code modifications on dedicated Git branches, run test and lint verification suites, and submit clean Pull Requests.
Jules includes four domain-specialized agents:
| Agent | Domain | Mission & Focus Areas |
|---|---|---|
| Sentinel 🛡️ | Security | Audits code for vulnerabilities: hardcoded secrets, SQL injection, XSS, CSRF, insecure endpoints, and data leaks. |
| Palette 🎨 | Design & UX | Identifies micro-UX and accessibility improvements: ARIA labels, focus states, keyboard navigation, missing loading/disabled states, empty states, and contrast. |
| Bolt ⚡ | Performance | Pinpoints bottlenecks and inefficiencies: N+1 queries, unoptimized loops, missing caches, unnecessary React re-renders, payload compression, and memory leaks. |
| Sweeper 🧹 | Maintainability | Cleans up technical debt and code smells: vague naming, magic numbers, complex boolean logic, bloated functions, duplicate code, and outdated syntax. |
All Jules agents operate via a deterministic Plan-and-Solve agentic lifecycle:
- Memory & Reflection: Prior to starting, the agent reads its long-term journal in
.jules/<agent>.md(if present) to recall past architectural decisions and avoid repeating pitfalls. - Analysis & Selection: Scans the workspace and isolates the single highest-priority issue that can be cleanly resolved in less than 50 lines.
- Branch Creation: Checks out a dedicated descriptive Git branch (e.g.,
perf/bolt-optimize-loopsorsec/sentinel-fix-csrf). - Implementation: Implements surgical, well-commented fixes.
- Verification: Executes workspace linters and test suites (
make lint,make test,pnpm test, etc.) to guarantee zero regressions. - Journal Reflection (Conditional): Appends non-obvious codebase learnings to
.jules/<agent>.mdonly if novel insights were discovered. - Atomic Commit & PR: Stages all changes in a single commit and creates a structured Pull Request.
Jules can be executed natively as a standalone binary or inside an isolated containerized workspace.
Ensure Go 1.25+ is installed:
# Build the binary to bin/
make build
# Run Sentinel to audit security
./bin/jules-ai-agent --agent sentinel --workspace /path/to/project --prompt "Audit codebase for security vulnerabilities"
# Run Bolt to optimize performance
./bin/jules-ai-agent --agent bolt --workspace /path/to/project --prompt "Identify and fix the most critical performance bottleneck"
# Run Palette for accessibility and UX
./bin/jules-ai-agent --agent palette --workspace /path/to/project --prompt "Improve keyboard navigation and ARIA accessibility"
# Run Sweeper for maintainability cleanup
./bin/jules-ai-agent --agent sweeper --workspace /path/to/project --prompt "Clean up technical debt and refactor bloated routines"Jules provides a pre-configured Ubuntu Noble development container packed with developer toolchains (Node.js 22, Python 3.12, Go, PHP 8.4, Docker-in-Docker, and systemd):
# Start the Jules workspace container
docker compose up -d --build
# Run an agent inside the isolated workspace
docker compose exec jules bash -c "jules --agent sentinel --workspace /workspace/project --prompt 'Audit for security vulnerabilities'"| Flag | Type | Description | Default |
|---|---|---|---|
--agent |
string | (Required) Agent to run: bolt, palette, sentinel, or sweeper |
"" |
--workspace |
string | Target workspace directory path to analyze | "." |
--prompt |
string | (Required) Task instructions and context prompt for the agent | "" |
--verbose |
bool | Enable verbose/debug logging output from the agent runtime | false |
- Go 1.25+: golang.org
- Make: Standard build automation tool
- Docker: For containerized workspace builds and verification
# Build binary to bin/
make build
# Run unit tests
make test
# Run unit tests with race detection and coverage
make test-coverage
# Format Go source code
make fmt
# Run linter
make lint
# Cross-compile for all supported operating systems and architectures
make build-all
# Build local Docker image
make docker-build
# Build multi-architecture Docker image with buildx
make docker-build-multiarch
# Clean build artifacts
make clean- Contributing Guide: Guidelines for local setup, pull requests, and Conventional Commits.
- Agent Guidelines: Agentic architecture, subagents, and memory journal conventions.
- Living Product Roadmap: Upcoming architectural items and feature plans.
- Code of Conduct: Contributor Covenant Code of Conduct.
- Security Policy: Vulnerability disclosure channels and SLAs.
This project is open-source software licensed under the Apache License, Version 2.0. See the LICENSE file for the complete license text.