A free, GitHub-native autonomous coding agent from Google that is quick to try but still short on independent proof of both capability and security.
Jules is Google's autonomous coding agent, built inside Google Labs and powered by Google's Gemini models. It connects to a GitHub repository, clones the code into an isolated cloud virtual machine, plans a set of changes, and opens a pull request for review. Google previewed Jules in December 2024, opened a public beta in May 2025, and moved it to general availability with free and paid tiers in August 2025.
Jules connects to a GitHub account and repository and can be assigned a task directly or through the 'jules' label on a GitHub issue.
Jules reads the codebase, including an optional AGENTS.md file, and produces an editable plan before making any changes.
The task runs in a dedicated Google Cloud virtual machine where Jules installs dependencies and modifies files; multiple tasks can run at once in separate VMs.
Jules opens a GitHub pull request with its changes, can respond to PR comments, and can attempt to fix failing CI checks automatically.
EDITORIAL NOTEApplied a small downward adjustment because two independent security disclosures in 2025 (prompt-injection data exfiltration and a remote-control exploit) were not confirmed fully remediated as of publication, a severity the privacy sub-score's 15% weight may understate for an agent with GitHub write access and code execution.
AI synthesis of external reviews · not on bestaiq
Synthesized from 3 external reviews. Independent signal (Trustpilot / Reddit / verified aggregators) weighted higher than commission-carrying review sites.
Yes. The free tier allows 15 tasks per day with 3 running at once, on Gemini 2.5 Pro. Paid plans raise those limits and give access to newer Gemini models.
Jules is included in Google AI Pro at $19.99 a month, which raises the limit to 100 tasks a day and 15 concurrent tasks. Google restructured its Ultra plan in May 2026 into a $100 a month tier and a $200 a month tier; which one carries Jules' full 300 tasks a day and 60 concurrent tasks has not been confirmed on an official page, so check current pricing directly before subscribing.
According to Jules' own documentation, it does not train on private repository code. If a connected repository is public, its data may be used for training. Google said an August 2025 wording update to this policy clarified the language without changing the underlying behavior.
Yes. An independent security researcher disclosed a prompt injection vulnerability in 2025 that could exfiltrate code to an outside server, and a separate issue where a malicious GitHub issue could lead Jules to download and run unauthorized software inside its task environment. Google was notified of both; the researcher said the underlying gaps were not confirmed fixed as of publication.
There is no rigorous, independently verified benchmark comparing Jules against these tools. Informal, single-author tests and forum comments exist but are not standardized; some found Jules behind competitors on task completion, and one found it added tests more consistently than Codex in early testing. Treat any specific win-rate or percentage figure for Jules against a named competitor as unverified unless it cites a primary source.
Jules is a genuinely free way to try an autonomous, GitHub-native coding agent from Google, and its toolset has grown quickly since 2024. It currently lacks an independently verified benchmark score, a security certification, and a documented data retention policy, and it has two disclosed 2025 security findings that were not confirmed fixed as of publication. Review every change carefully before merging, and confirm current pricing on jules.google before subscribing, since the Ultra plan's price and task limits changed in 2026.