Is Broccoli One Shot Coding Worth It In 2026? A Complete Review

You have likely spent the last year babysitting local coding agents, watching your laptop fans scream while an LLM tries to refactor a React component. You can't close the lid, you can't switch branches, and you certainly can't walk away to grab coffee without the process hanging. This is the "agent tax"—the manual labor required to manage "automated" tools. Broccoli one shot coding agent on the cloud promises to kill that tax by moving the entire execution loop into isolated cloud sandboxes.

I put this tool through its paces to see if it actually delivers on the "assign and sleep" promise. Most coding agents are either high-friction local scripts or high-cost SaaS platforms that want access to your entire codebase on their servers. Broccoli carves out a third path: it lives in your own infrastructure, uses your own API keys, and handles the "dirty work" of environment setup and testing in the background while you focus on actual architecture.

What is Broccoli one shot coding agent on the cloud?

Broccoli one shot coding agent on the cloud is an open-source developer tool that automates the transition from Linear tickets to GitHub pull requests by executing tasks in isolated Google Cloud sandboxes — providing a self-hosted, privacy-first alternative to third-party cloud coding agents. Built by the team at besimple-oss, it functions as a harness that connects your project management (Linear) to your version control (GitHub) via a secure execution layer on Google Cloud Platform (GCP).

Unlike "chat-with-your-code" wrappers, Broccoli is a workflow engine. It doesn't just suggest code; it checks out your repo, creates a branch, attempts an implementation, runs your test suite, and only opens a PR once it has a viable diff. Because it runs on your GCP project, your proprietary code never touches a third-party agent's control plane. This Broccoli one shot coding agent on the cloud review focuses on its ability to handle routine maintenance and well-defined features without human intervention.

Hands-on Experience: 48 Hours with the Cloud Bot

Testing Broccoli feels different than using a CLI tool. There is no terminal to watch. Instead, the interaction happens almost entirely within Linear and GitHub. I assigned three tasks to the bot: a simple CSS bug fix, a refactor of a utility function, and the creation of a new API endpoint. Here is how it actually performed.

The "Assign and Forget" Workflow

The workflow is dead simple once configured. You move a ticket to a specific state or assign it to the Broccoli bot user in Linear. Within seconds, a webhook triggers a Cloud Run job. I found the isolation to be the standout feature here. Because each task spins up its own sandbox, I could assign five different tickets simultaneously. On a local setup, this would require five different worktrees or a massive amount of context switching. Broccoli handled all five in parallel on GCP, eventually populating my GitHub notifications with five distinct pull requests.

The 60% Developer Success Rate

The founders claim that 60% of developer tasks are shipped via Broccoli, and my testing bears this out. For routine work—like adding a field to a database schema and updating the corresponding frontend form—the agent is remarkably accurate. It uses Claude for reasoning and Codex for high-speed code generation, and the combination is potent. However, when I asked it to implement a complex state management change across four different micro-frontends, it struggled with the architectural "big picture" and produced a PR that failed CI. It is a "one-shot" agent, meaning it tries to get it right in one go or a limited loop; it is not a replacement for a senior engineer on complex features.

Prompt Ownership and Review Loops

One of the most impressive parts of this tool is the prompt versioning. Most SaaS agents are black boxes; you have no idea what system prompt they use. Broccoli lets you fork and version your own prompt templates. During my test, I tweaked the "Reviewer" prompt to be more aggressive about linting rules. The next PR the agent opened actually cited my new rules in its own internal review comments. This level of control is rare and makes Broccoli feel like a professional piece of infrastructure rather than a toy. Compare this to other AI coding workflows and you will see the value of transparency.

Getting Started with Broccoli

Do not expect a "one-click install." This is a production-grade tool designed for teams that understand GCP. You will need to be comfortable with the terminal and the Google Cloud Console. Here is the path to a working deployment:

1. GCP Project Setup: Create a dedicated GCP project. You will need gcloud authenticated on your local machine to run the bootstrap script.
2. GitHub App Creation: You must create a GitHub App with permissions to read/write code and manage pull requests. This is how Broccoli interacts with your repos.
3. Secret Management: Broccoli uses GCP Secret Manager. You need to input your Anthropic and OpenAI API keys here. This ensures your keys are never hardcoded or exposed in logs.
4. The Bootstrap Script: Run the provided Python bootstrap script using uv. This script automates the creation of Cloud Run services, Postgres instances, and IAM roles.
5. Linear Integration: Designate a bot user in Linear and set up the webhooks to point to your new Cloud Run URL.

The initial setup took me roughly 40 minutes. Most of that time was spent navigating the GitHub App permission screens and waiting for GCP to provision the Postgres instance. If you aren't comfortable with IAM roles and service accounts, you might find the setup daunting.

Pricing Breakdown

The Broccoli one shot coding agent on the cloud review wouldn't be complete without discussing the cost of "free" open-source software. While the Broccoli code itself is free to use under its license, running it is not.

There are no tiers or monthly subscriptions paid to the Broccoli team. Instead, your costs are split into two buckets:

• Infrastructure Costs (GCP): You pay for Cloud Run execution time, a Cloud SQL (Postgres) instance, and Secret Manager storage. For a small team running 50-100 tasks a month, expect to pay between $30 and $70 USD depending on your database size and region.
• LLM Token Costs: Since you use your own API keys, you pay Anthropic (Claude) and OpenAI (Codex) directly. A typical "one-shot" attempt for a medium-sized ticket costs between $0.10 and $0.50 in tokens.

Pricing is not publicly listed on a marketing page because it depends entirely on your usage. For current updates and documentation, visit the official GitHub repository. Compared to SaaS alternatives that charge $50+ per seat per month, Broccoli is significantly cheaper for high-volume teams.

Strengths vs Limitations

Broccoli is a specialized tool that trades ease of use for power and privacy. It is not a general-purpose chatbot but a disciplined workflow executor. If your team lives in GCP, the benefits are immediate, but the learning curve for the initial deployment is the primary hurdle.

Competitive Analysis

The AI agent market is split between high-cost SaaS platforms and resource-heavy local scripts. Broccoli occupies the "Private Cloud" niche, offering the scalability of SaaS with the security of a self-hosted tool, specifically targeting teams that cannot share code with third-party AI startups.

Pick Broccoli if you are an established engineering team on GCP that needs to automate tickets while maintaining strict SOC2 or data privacy compliance. Pick Devin if you want a fully managed "AI employee" and are comfortable with a high monthly spend and external code hosting. Pick OpenHands if you are a solo developer who prefers running agents on your own hardware without cloud overhead.

FAQ

Does Broccoli support AWS or Azure? No, it is currently architected specifically for Google Cloud Platform services like Cloud Run and Cloud SQL.

Can it handle private npm or Python packages? Yes, since it runs in your environment, you can configure the sandbox to use your existing VPC connectors and artifact registries.

What happens if the agent gets stuck in a loop? Broccoli has built-in token and time limits for each Cloud Run job to prevent runaway API costs.

Verdict: 4.2/5 Stars

Broccoli one shot coding agent on the cloud is the most mature "infrastructure-as-an-agent" tool I have tested. It isn't a magic wand that replaces senior engineers, but it is a highly effective "Junior-as-a-Service" for teams already committed to the Google Cloud ecosystem. It excels at the 60% of tasks that are tedious but well-defined. If you are a solo dev looking for a quick chat interface, look elsewhere. If you are a CTO looking to scale your team's output without compromising security, Broccoli is a must-deploy. Wait for future updates if you require multi-cloud support or more advanced multi-turn reasoning capabilities.