Coding Horizon

10 Free Coding Agents That Replace Paid AI Tools

Every figure, date, licence and claim shown in this video, with the primary source it was taken from. Checked 9 August 2026.

What the paid tools cost

Product Plan Price
Cursor Hobby Free
Cursor Individual Pro $20 / month
Cursor Teams Standard $40 / user / month
Claude Pro $20 / month billed monthly ($17 / month billed annually)
Claude Max From $100 / month
GitHub Copilot Free $0
GitHub Copilot Pro $10 / month
GitHub Copilot Pro+ $39 / month
GitHub Copilot Max $100 / month

GitHub Copilot’s paid plans include a monthly allowance of GitHub AI Credits — $15 on Pro, $70 on Pro+ and $200 on Max — after which usage continues as paid usage against a budget the customer sets. One AI credit is $0.01. Copilot Free is limited to 2,000 completions and 50 chat requests.

Claude Code is included on Claude’s Pro and Max plans. The pricing page states that usage limits apply but publishes no per-plan figure for them, so no number for those limits is shown in this video.

10. mini-SWE-agent

Described by its own project as “Minimal: Just some 100 lines of python for the agent class”, with no heavy dependencies. It works with models through litellm, OpenRouter, Portkey and others, so the model is supplied separately and can be local.

SWE-agent, the larger project it came out of, “enables your language model of choice … to autonomously use tools to fix issues in real GitHub repositories”. It is MIT licensed and was started at Princeton University, with continuing work from Princeton and Stanford. Its own documentation now carries the recommendation to use mini-SWE-agent instead: “Our general recommendation is to use mini-SWE-agent instead of SWE-agent going forward”, and mini-SWE-agent’s own README says “You should consider mini-swe-agent your default choice.”

9. Continue

The repository banner reads: “The continuedev/continue repository is no longer actively maintained and is read-only for all users.” The project shipped a final 2.0.0 release covering the original VS Code extension, the CLI and the JetBrains plugin, which included removing anonymous telemetry, pulling out authentication and bug fixes. The code remains available under Apache 2.0, marked “Apache 2.0 © 2023-2026 Continue Dev, Inc.”

8. Qwen Code

Alibaba’s open-source terminal coding agent, from the QwenLM organisation. It connects to OpenAI, Anthropic, Gemini and Qwen APIs as well as third-party providers and local models through Ollama or vLLM, and can switch between them at runtime.

Its authentication documentation states: “The Qwen OAuth free tier was discontinued on 2026-04-15. Existing cached tokens may continue working briefly, but new requests will be rejected.” The remaining options are the Alibaba Cloud Coding Plan, built-in third-party providers including DeepSeek, MiniMax and OpenRouter, or a custom provider pointed at a local server or proxy.

7. Kilo Code

Runs across “VS Code, JetBrains, and the CLI”, offers “500+ models” with the ability to “switch between them mid-task”, and ships several specialised agents — Code, Plan, Ask, Debug and Review — that can be switched depending on the task, alongside terminal and browser control. MIT licensed.

Its sandbox is described as “OS-native process confinement applied to every tool the agent executes — including tools that already passed permission checks”, restricting writes to the workspace and approved paths, keeping .git read-only, and optionally forcing network traffic through an allowlisted proxy.

6. Aider

Lets you “pair program with LLMs to start a new project or build on your existing codebase”. It “makes a map of your entire codebase, which helps it work well in larger projects”, edits multiple files, and “works with most popular programming languages: python, javascript, rust, ruby, go, cpp, php, html, css, and dozens more”. It “automatically commits changes with sensible commit messages” and lets you “use familiar git tools to easily diff, manage and undo AI changes”. It runs against “Cloud and local LLMs” and works with “almost any LLM, including local models”, which is what makes a fully local setup with Ollama possible.

5. Goose

The canonical repository sits in Block’s GitHub organisation, and the project states that “goose is part of the Agentic AI Foundation (AAIF) at the Linux Foundation”. Apache 2.0 licensed. It ships as “A native desktop app for macOS, Linux, and Windows. A full CLI for terminal workflows. An API to embed it anywhere.”

It connects to “70+ extensions via the Model Context Protocol open standard” and works with “15+ providers — Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure, Bedrock, and more”. It is explicitly broader than coding: “Not just for code — use it for research, writing, automation, data analysis, or anything you need to get done.”

4. Cline

Cline “reads your project structure, understands the relationships between files, and makes coordinated changes across your codebase”, “executes commands directly in your terminal and watches the output in real time”, and can “create files, run commands, browse the web, and use tools with human-in-the-loop approval”. Beyond the original VS Code extension it offers a CLI (“Interactive chat or fully headless for CI/CD and scripting”), a web task board to “Run many agents in parallel”, and Multi-Agent Teams. It is “not locked to a single AI provider”, supporting Claude, GPT, Gemini, Ollama, local models and “Any OpenAI-compatible API”. Apache 2.0 licensed.

3. OpenHands

Describes itself as “The self-hosted developer control center for coding agents and automations”. It offers a web interface, a CLI and an SDK through its Agent Server REST API. On models it says “Bring your own model” and “Use with any LLM”, and on agents “Use with any agent” — OpenHands, Claude Code, Codex, Gemini, or anything speaking the Agent-Client Protocol. The self-hosted version is open source and MIT licensed.

2. Gemini CLI

Google’s official repository documents a free tier of “60 requests/min and 1,000 requests/day with personal Google account”. The CLI itself is Apache 2.0 licensed.

Google Cloud’s own quota documentation lists a higher figure — 1,500 requests per user per day — for the paid Standard edition used in agent mode or through Gemini CLI. That is a different, paid tier, not the free personal-account allowance, and the figure shown in this video is the free-tier one from the Gemini CLI repository.

1. OpenCode

“OpenCode is an open source AI coding agent. It’s available as a terminal-based interface, desktop app, or IDE extension.”

Its stated capabilities: “Any model — 75+ LLM providers through Models.dev, including local models”; “LSP enabled — Automatically loads the right LSPs for the LLM”; “Multi-session — Start multiple agents in parallel on the same project”. It also supports signing in with existing subscriptions: “Log in with GitHub to use your Copilot account” and “Log in with OpenAI to use your ChatGPT Plus or Pro account”.

Not checked