Best AGENTS.md alternatives (2026)

What it is: goose is the only agent in this set governed by a foundation rather than a vendor, and it can borrow an existing Claude, ChatGPT or Gemini subscription over the Agent Client Protocol instead of an API key. [github.com/aaif-goose/goose]

Choose Goose when: Automate work that is not coding — research, writing, data analysis — on the same machine and the same agent, through the desktop app, the CLI or the API. [github.com/aaif-goose/goose]

Where it hurts: Model choice is not free: goose leans on tool calling, so a weak or local model degrades it, and the vendor names Claude 4 as the model it works best with. [goose-docs.ai/docs/getting-started]

Price: goose charges nothing for itself; you pay whichever provider you configure, and it accepts a ChatGPT Plus or Pro subscription in place of an API key. [goose-docs.ai/docs/getting-started]

What it is: Qwen Code binds the local agent to Telegram, DingTalk, WeChat or Feishu with one command, `qwen channel`, and ships TypeScript, Python and Java SDKs beside the CLI. [github.com/QwenLM/qwen-code]

Choose Qwen Code when: Run the agent from a group chat your team already uses, with `qwen channel`. [github.com/QwenLM/qwen-code]

Where it hurts: Qwen Code is no longer free: the Qwen OAuth free tier was discontinued on 2026-04-15, and requests on an old token are now rejected. [qwenlm.github.io/qwen-code-docs/en]

Price: The free tier is gone as of 2026-04-15; the vendor's recommended replacement is the Alibaba ModelStudio Coding Plan, whose weekly quota is priced on Alibaba's own page, which I did not read today. [qwenlm.github.io/qwen-code-docs/en]

What it is: OpenCode reaches 75+ LLM providers and local models from a single config file, because it resolves every model through the AI SDK and the Models.dev catalogue instead of a hand-written provider list. [opencode.ai/docs/providers]

Choose opencode when: Explore an unfamiliar codebase before you change it: the built-in plan agent refuses file edits and asks before each bash command. [github.com/anomalyco/opencode]

Where it hurts: A fresh install edits files and runs shell commands with no approval prompt, because every permission except `doom_loop` and `external_directory` starts at allow. [opencode.ai/docs/permissions]

Price: OpenCode Go, the optional subscription, costs $5 for the first month and $10 a month after that, and one workspace can hold only one subscriber. [opencode.ai/docs/go]

What it is: Crush is configured by a `crushrc` that is Bash, and it runs that Bash identically on Windows because Crush embeds its own Bash interpreter rather than calling the host shell. [github.com/charmbracelet/crush]

Choose Crush when: Code in a terminal on a platform the other agents skip: Crush ships first-class builds for Android, FreeBSD, OpenBSD and NetBSD as well as macOS, Linux and Windows. [github.com/charmbracelet/crush]

Where it hurts: Crush is not open source today: it ships under the Functional Source License 1.1, which bars any competing commercial use until each version turns MIT two years after release. [github.com/charmbracelet/crush]

Price: Charm Hyper, the provider built for Crush, gives 100 Hypercredits a month at $0 and 250 a day on a $20 monthly subscription; 1 Hypercredit is 5 cents. [hyper.charm.land]

What it is: Claude Code runs only Anthropic models: every alias — default, best, fable, sonnet, opus, haiku and opusplan — resolves to a Claude model, and changing the base URL moves where the request goes without changing which model answers it. [code.claude.com/docs/en]

Choose Claude Code when: Plan a change with Opus and execute it with Sonnet inside one session, by setting the model to `opusplan`. [code.claude.com/docs/en]

Where it hurts: You cannot audit, fork or patch the agent: the repository publishes the changelog, plugins and examples, not the CLI source, under an all-rights-reserved licence. GitHub Copilot CLI is the only other tool in this set that does the same. [github.com/anthropics/claude-code]

Price: Claude Pro costs $17 per month on the annual plan, $200 billed up front, or $20 per month billed monthly; Max starts at $100 per month. [claude.com/pricing]

What it is: Google Antigravity is a platform rather than an editor: its IDE extensions put the same agent — side-panel chat, inline diff review and interactive plans — inside the editor you already use, so adopting Antigravity does not require adopting its IDE. [antigravity.google/docs/ide]

Choose Google Antigravity when: Run Antigravity without leaving Visual Studio Code: Google's own extension puts agent chat, inline diff review and interactive plans into VS Code side panels, and installs a local `agy` backend service the first time you launch it. [antigravity.google/docs/ide]

Where it hurts: You cannot point Antigravity at a model key you already hold: the plans page states there is no bring-your-own-key or bring-your-own-endpoint path, so the quota Google grants you is your only supply. [antigravity.google/docs/plans]

Price: Antigravity costs $0 a month for an individual, and the pricing page now labels that tier Generally Available rather than a preview. [antigravity.google/pricing]

What it is: Vibe Kanban is an orchestrator, not an agent: each kanban issue gets its own branch, terminal and dev server, and the board drives 10 named coding agents including Claude Code, Codex, Copilot and OpenCode. [github.com/BloopAI/vibe-kanban]

Choose vibe-kanban when: Compare what two agents did to the same issue and comment on the diff without leaving the board. [github.com/BloopAI/vibe-kanban]

Where it hurts: The company behind it is gone: bloop announced its shutdown on 2026-04-10 because it could not find a business model. [vibekanban.com/blog/shutdown]

Price: Paid subscriptions were cancelled and refunded when bloop shut down on 2026-04-10; the local product now costs nothing. [vibekanban.com/blog/shutdown]