Last checked November 5, 2026
Free alternatives to Jules
Jules is Google's asynchronous coding agent. The homepage says you pick a GitHub repository and a branch, write a prompt, and Jules opens a pull request. A jules label on a GitHub issue assigns a task. The October 29, 2026 changelog says you can finish sessions through November 11, 2026. After that date Jules is read-only, and it winds down on December 31, 2026. The homepage checked the same day does not state those dates. The changelog says this is part of unifying Google's agentic coding efforts. Those lessons are shaping Google Antigravity. Code, branches, and pull requests already in GitHub stay there. The export is the settings and history that live only inside Jules. The price tracker records price changes. A shutdown with no dollar figure stays on this page.
- Other developer tools on this site. DevOps
- Export the Jules folder, then start on the closest app. Move from Jules to OpenHands
- The closest app on this list. OpenHands
- A terminal and desktop agent that the docs show opening a pull request. OpenCode
- Git commits from the terminal, with a quiet release cadence. Aider
- Code, branches, and pull requests already in GitHub stay there. Free alternatives to GitHub
- Sourcegraph Cody reaches end of life on January 31, 2027. Free alternatives to Sourcegraph Cody
- A paid editor scored on the Sourcegraph Cody checklist, not this one. Cursor
What you pay for Jules
The homepage lists task caps and does not print a dollar price. Jules winds down on December 31, 2026.
Checked November 5, 2026. The homepage lists a Jules plan at 15 tasks a day and 3 concurrent tasks on Gemini 2.5 Pro, Jules in Pro at 100 tasks a day and 15 concurrent tasks, and Jules in Ultra at 300 tasks a day and 60 concurrent tasks, both starting with Gemini 3 Pro. It does not print a monthly dollar price. We have not checked one. The October 29, 2026 changelog is the shutdown notice: finish sessions through November 11, 2026, read-only after that until year end, and wind down on December 31, 2026.
Who it is for
For developers, individuals, and small teams. Its price on file: The homepage lists task caps and does not print a dollar price. Jules winds down on December 31, 2026. Not checked. It is in the DevOps group.
How to switch
The closest free app on the checklist is OpenHands. Cost and details. Move from Jules to OpenHands.
An agent you can still run
These scores are from the checklist on this page. A full score means it matches our checklist. It is not a drop-in replacement. We check each app against public docs and price pages.
OpenHands is #1 (84). OpenCode is #2 (78). Aider is #3 (45).
OpenHands
#1 · Closest
8 of 8 checked features. A full score means it matches our checklist. It is not a drop-in replacement.You only get part of: opens a branch or a pull request, environment variables for the task, and start a task from a GitHub issue.
OpenCode
#2
8 of 8 checked features. A full score means it matches our checklist. It is not a drop-in replacement.You only get part of: a session you can leave and come back to, a task that runs on a schedule, and notes the agent keeps for the next task, and 1 more.
Aider
#3
8 of 8 checked features. A full score means it matches our checklist. It is not a drop-in replacement.What you give up: a task that runs on a schedule, environment variables for the task, and start a task from a GitHub issue. You only get part of: a session you can leave and come back to, works with a GitHub repository, and opens a branch or a pull request, and 1 more.
Where they differ
A session you can leave and come back to
Worth 8 points. OpenHands does all of it. The README says agents on a server keep running after the laptop is shut. The Docker sandbox notes say conversation history survives a container replacement. OpenCode does part of it. The homepage says you can start multiple agents in parallel on one project. That is a local session. The docs checked do not describe a cloud session that keeps working after you close the laptop. Aider does part of it. The git docs describe a terminal chat with /undo. The scripting docs run one message and exit. They do not describe a session that keeps running on a server after you leave.
Works with a GitHub repository
Worth 8 points. OpenHands does all of it. The PR review guide connects a GitHub token and a GitHub MCP server, and automations can name a repository. OpenCode does all of it. The GitHub docs install an app and a workflow so OpenCode runs inside GitHub Actions on a repository. Aider does part of it. The git docs say aider works inside a git repository, which can be a GitHub clone. The GitHub page is about using Copilot as a model API, not a connection to your repositories.
Opens a branch or a pull request
Worth 8 points. OpenHands does part of it. The documented GitHub workflow reviews an existing pull request and can be given pull request write permission. It does not say Agent Canvas opens a new pull request for every task. OpenCode does all of it. The same docs say /opencode on an issue can create a branch and open a pull request. A review comment can commit onto the current pull request. Aider does part of it. Each edit is committed, and the git docs say you can keep a series of changes on a branch. They do not describe opening a pull request.
A task that runs on a schedule
Worth 6 points. OpenHands does all of it. The automations guide shows a schedule trigger, including a cron expression, plus webhook and event triggers. OpenCode does part of it. A documented GitHub Actions event runs OpenCode on a cron schedule. That is a workflow file in the repo, not a scheduled task stored in the app. Aider does none of it. The docs index checked on November 5, 2026 has scripting and a chat. It does not have a scheduler. A shell loop in the scripting doc is something you start yourself.
Notes the agent keeps for the next task
Worth 5 points. OpenHands does all of it. Settings include persistent agent memory. When it is on, new conversations can load saved memory from the workspace and the user. OpenCode does part of it. The intro's /init command writes AGENTS.md and tells you to commit it. That is a file you keep. It is not a memory Jules saved for you. Aider does part of it. The conventions doc says you write CONVENTIONS.md and load it with the chat, including from .aider.conf.yml. That file is yours. Aider does not extract memories from old tasks.
Environment variables for the task
Worth 5 points. OpenHands does part of it. Settings store secrets for the active backend. The docs checked do not describe a per-task environment variable list like the Jules export. OpenCode does part of it. The GitHub setup stores a provider API key as an Actions secret. The docs checked do not describe a saved list of task environment variables. Aider does none of it. The scripting page lists environment variables for aider's own flags, such as AIDER_MESSAGE. It does not describe environment variables for the project task.
Start a task from a GitHub issue
Worth 5 points. OpenHands does part of it. The README says automations can decompose GitHub issues into tasks. It does not describe a label that assigns one issue to the agent. OpenCode does all of it. Comment /opencode on an issue, or run the documented issues workflow when an issue is opened. The fix example opens a pull request from that issue. The page is labeled as the V1 docs. Aider does none of it. The docs checked do not start a task from a GitHub issue. The GitHub page is a Copilot model setup.
OpenHands, OpenCode, and Aider do all of these: an agent edits the repo from a prompt.
Google Antigravity
No scoreThe agent platform named in the Jules shutdown note. Not scored on this checklist.
Claude Code
No scoreA terminal coding agent from Anthropic. Not scored on this checklist.
How close is it?
How this worksA full score means it matches our checklist. It is not a drop-in replacement. #1 is the closest free app that did not pay us. An ad, when one is sold, is marked and sits after #1. Paying does not change the score. Open source means you can use, change, and share the code. Run it yourself means the app can live on a computer you control.
- #1 · ClosestOpen source
OpenHands
Agent Canvas, a self-hosted control center for coding agents.
8 of 8 checked features. A full score means it matches our checklist. It is not a drop-in replacement.You only get part of: opens a branch or a pull request, environment variables for the task, and start a task from a GitHub issue.
Last checked November 5, 2026 · License: MIT
MacWindowsLinuxWebRun it yourself - #2Open source
OpenCode
A terminal and desktop coding agent that can open a GitHub pull request.
8 of 8 checked features. A full score means it matches our checklist. It is not a drop-in replacement.You only get part of: a session you can leave and come back to, a task that runs on a schedule, and notes the agent keeps for the next task, and 1 more.
Last checked November 10, 2026 · License: MIT
MacWindowsLinuxiOSWebRun it yourself - #3Open source
Aider
Terminal pair programming that commits each edit to git.
8 of 8 checked features. A full score means it matches our checklist. It is not a drop-in replacement.What you give up: a task that runs on a schedule, environment variables for the task, and start a task from a GitHub issue. You only get part of: a session you can leave and come back to, works with a GitHub repository, and opens a branch or a pull request, and 1 more.
Last checked November 5, 2026 · License: Apache-2.0
MacWindowsLinux
Feature by feature
Full means it does the whole job. Partial means it does some of it. Missing means it does not. Not checked counts as zero. The number under each feature is how much it counts. A bigger number moves the score more.
Scroll sideways to see every app.
| Feature | OpenHands#1 · 848 of 8 checked features | OpenCode#2 · 788 of 8 checked features | Aider#3 · 458 of 8 checked features |
|---|---|---|---|
| An agent edits the repo from a promptCounts for 10 | FullThe README describes conversations where an agent works in a project. The setup docs say the agent can run shell commands and edit files inside its workspace. | FullThe intro docs say you describe a change and OpenCode edits the project. Plan mode can suggest the change before Build mode writes it. | FullThe scripting docs take one natural-language message, edit the files, and commit. The git docs describe the same edits from an interactive chat. |
| A session you can leave and come back toCounts for 8 | FullThe README says agents on a server keep running after the laptop is shut. The Docker sandbox notes say conversation history survives a container replacement. | PartialThe homepage says you can start multiple agents in parallel on one project. That is a local session. The docs checked do not describe a cloud session that keeps working after you close the laptop. | PartialThe git docs describe a terminal chat with /undo. The scripting docs run one message and exit. They do not describe a session that keeps running on a server after you leave. |
| Works with a GitHub repositoryCounts for 8 | FullThe PR review guide connects a GitHub token and a GitHub MCP server, and automations can name a repository. | FullThe GitHub docs install an app and a workflow so OpenCode runs inside GitHub Actions on a repository. | PartialThe git docs say aider works inside a git repository, which can be a GitHub clone. The GitHub page is about using Copilot as a model API, not a connection to your repositories. |
| Opens a branch or a pull requestCounts for 8 | PartialThe documented GitHub workflow reviews an existing pull request and can be given pull request write permission. It does not say Agent Canvas opens a new pull request for every task. | FullThe same docs say /opencode on an issue can create a branch and open a pull request. A review comment can commit onto the current pull request. | PartialEach edit is committed, and the git docs say you can keep a series of changes on a branch. They do not describe opening a pull request. |
| A task that runs on a scheduleCounts for 6 | FullThe automations guide shows a schedule trigger, including a cron expression, plus webhook and event triggers. | PartialA documented GitHub Actions event runs OpenCode on a cron schedule. That is a workflow file in the repo, not a scheduled task stored in the app. | MissingThe docs index checked on November 5, 2026 has scripting and a chat. It does not have a scheduler. A shell loop in the scripting doc is something you start yourself. |
| Notes the agent keeps for the next taskCounts for 5 | FullSettings include persistent agent memory. When it is on, new conversations can load saved memory from the workspace and the user. | PartialThe intro's /init command writes AGENTS.md and tells you to commit it. That is a file you keep. It is not a memory Jules saved for you. | PartialThe conventions doc says you write CONVENTIONS.md and load it with the chat, including from .aider.conf.yml. That file is yours. Aider does not extract memories from old tasks. |
| Environment variables for the taskCounts for 5 | PartialSettings store secrets for the active backend. The docs checked do not describe a per-task environment variable list like the Jules export. | PartialThe GitHub setup stores a provider API key as an Actions secret. The docs checked do not describe a saved list of task environment variables. | MissingThe scripting page lists environment variables for aider's own flags, such as AIDER_MESSAGE. It does not describe environment variables for the project task. |
| Start a task from a GitHub issueCounts for 5 | PartialThe README says automations can decompose GitHub issues into tasks. It does not describe a label that assigns one issue to the agent. | FullComment /opencode on an issue, or run the documented issues workflow when an issue is opened. The fix example opens a pull request from that issue. The page is labeled as the V1 docs. | MissingThe docs checked do not start a task from a GitHub issue. The GitHub page is a Copilot model setup. |
What it costs
OpenHands
- Free plan
- The README's install is npm or Docker, and the LICENSE file is MIT. OpenHands Cloud is a commercial option. We have not checked its price.
- They host it
- OpenHands Cloud is named in the README. We have not checked a dollar price.
- Run it yourself
- Setup: medium. The README's npm path needs Node.js 24 or later and uv. The Docker example checked on November 5, 2026 is tagged 1.35.0. The same README warns that an install without a sandbox can read and write files on that machine.
OpenCode
- Free plan
- The homepage says you can start with free models and no account. Models named on November 10, 2026 include Big Pickle, DeepSeek V4.1 Flash Free, and GLM-5.3-Flash Free.
- OpenCode Go is $10 a month. OpenCode Go Plus is $40 a month.
- A ChatGPT Plus or Pro login, or a GitHub Copilot login, is a separate bill.
- They host it
- Free models are $0. OpenCode Go is $10 a month. OpenCode Go Plus is $40 a month. Checked November 10, 2026. Bringing ChatGPT or GitHub Copilot uses that subscription.
- Run it yourself
- Easy. The homepage lists desktop builds for Mac, Windows, and Linux, and a terminal install via a shell script, npm, or Homebrew. This listing did not run it.
Aider
- Free plan
- The Apache-2.0 license and the PyPI package do not list a fee for Aider. You pay the model provider. We have not checked those prices.
- They host it
- No hosted price on file.
- Run it yourself
- Easy. The GitHub Copilot doc says to install with pip and aider-install, then run aider in a project directory. It runs on the computer where you installed it.
Did you switch?
If you left Jules, tell us if it worked and how long it took. Notes stay private until we approve them. Nothing is shown until at least five are approved.
Closest pair
Move from Jules to OpenHands
Export Jules to a folder on your computer, then start Agent Canvas on the GitHub repo you already have. Nothing in the OpenHands docs reads that folder.
Read how to switchRelated alternatives
- Free alternatives to Bitbucket
- Free alternatives to Conductor
- Free alternatives to GitHub
- Free alternatives to Ollama Cloud
- Free alternatives to Ona
- Free alternatives to Sentry
Switch guides
Free apps in this group
- Aider: Terminal pair programming that commits each edit to git.
- Bugsink: A smaller Sentry-compatible error store.
- Cline: An Apache-2.0 coding agent for VS Code and the terminal. No inline autocomplete.
- Continue: An Apache-2.0 IDE assistant. The repository is read-only after the Cursor acquisition.
- cState: A Hugo status page. Incidents are markdown files.
- Cursor: A coding editor with Tab autocomplete and an agent. Free is limited. The code is closed.
- Devin: The desktop editor formerly sold as Windsurf. Free includes Tab. The code is closed.
- Errbit: An older Airbrake-compatible error catcher.
- Forgejo: A community forge, GPL-3.0 since v9.
- Gatus: A health dashboard with a status page in the config file.
- Gitea: A lightweight Git forge.
- GitHub Copilot: GitHub's coding assistant. Free is capped. The code is not public.
- GitLab Community Edition: The MIT edition of GitLab, without EE.
- GlitchTip: Error tracking compatible with Sentry SDKs.
- Highlight: Session replay and errors, with a split license.
- Kener: A status page with incidents, maintenance, and email subscriptions.
- llama.cpp: LLM inference in C and C++. The license is MIT. llama-server speaks an OpenAI-style API.
- Ollama: Run open models on your own computer. The license is MIT.
- OneUptime: Status pages plus monitors, incidents, and on-call.
- OpenCode: A terminal and desktop coding agent that can open a GitHub pull request.
- OpenHands: Agent Canvas, a self-hosted control center for coding agents.
- Radicle: A peer-to-peer code forge.
- SigNoz: Traces, metrics, and logs in one app.
- SourceHut: Git hosting composed of small services.
- Tabby: A self-hosted completion server. Community is free for up to 5 users.
- Uptime Kuma: Uptime checks, with a public status page of those checks.
- vLLM: An Apache-2.0 model server with an OpenAI-compatible API.