Crush Review 2026: Charm's Glamorous Terminal AI Coder
4.4/ 5
What Is Crush?
Crush is a terminal-native AI coding assistant built by Charm Bracelet, the team behind Bubbletea, Bubbles, and the wider Charm ecosystem. It's written in Go and offers a visually polished TUI that wraps agentic LLM interactions directly into your command line. As of 2026, Crush sits at 26,605 GitHub stars and continues to attract developers who want a pretty, keyboard-driven alternative to GUI-based AI coding tools.
Charm TUI Pedigree
If you've used Bubbletea or any Charm component, you know the level of polish they bring to terminal interfaces. Crush inherits that DNA: smooth animations, a composable layout, and full mouse support in terminals that allow it. The whole experience feels like a native desktop app—but inside your terminal. That pedigree is Crush's biggest differentiator in a crowded CLI agent space.
Pricing & Model Support
Crush is completely free and open source (MIT license). There's no paid tier, no usage caps, and no telemetry. You bring your own API keys for the underlying LLMs. The tool supports dozens of providers including OpenAI, Anthropic, Google, and any OpenAI-compatible endpoint. Example 2024 (snapshot) per-token costs:
- openai/o1-pro: $150/M in, $600/M out
- anthropic/claude-opus-4.7-fast: $30/M in, $150/M out
- openai/gpt-5.5-pro: $30/M in, $180/M out
- openai/o3-pro: $20/M in, $80/M out
- anthropic/claude-opus-4: $15/M in, $75/M out
Workflow & Features
Terminal-First UX
Launch Crush with crush in your project directory. You get a split-pane interface: left is a chat/follow-up area, right is a real-time diff of file changes. You can switch between panes with Tab, select context files with Ctrl+Space, and scroll output with PgUp/PgDn. Every action is key-bindable; the help menu (?) is instantly available.
LSP Integration
Crush doesn't ship its own LSP client, but it leverages your editor's LSP or external language servers via the MCP protocol. By connecting to a local LSP endpoint, Crush can fetch diagnostics, type information, and symbol references. This grounds generated code in real-time project context—a step beyond naive prompt-wrapping.
MCP (Model Context Protocol)
Crush natively supports the Model Context Protocol, allowing you to plug in external tools: file system access, shell execution, web search, or custom APIs. The default install includes a fs_read and bash_exec tool. You can add your own via a mcp.json config. This makes Crush extensible beyond plain code generation.
GitHub Momentum
Crush crossed 26,605 stars in 2025–2026, with a active pull request community. The repo receives weekly updates, and the issue tracker shows responsive maintainers. While still younger than projects like OpenCode, the growth rate suggests strong adoption among terminal lovers.
Verdict
Crush is an excellent choice if you live in the terminal, prefer a beautiful TUI over web UIs, and want the freedom to choose any LLM backend. It's less suited for developers who want a turnkey managed service or need integrated IDE-level debugging. For terminal die-hards and Charm fans, Crush delivers the most glamorous AI coding experience on the command line.
Pros & Cons
- Pros:
- Stunning terminal UI inherited from Charm ecosystem
- Free and open source with no usage restrictions
- BYO-key: full choice of models and providers
- MCP extensibility for custom tooling
- Active community and polished DX
- Cons:
- Steep learning curve for developers new to terminal TUIs
- Requires manual API key setup—no built-in model credits
- Smaller plugin ecosystem compared to mature CLI agents
FAQ
Is Crush free to use?
Yes. Crush is free and open source. You only pay for the API tokens consumed by the LLMs you configure.
Can I use Crush without an API key?
No. Crush itself contains no built-in models; you must provide your own API keys for a supported provider.
How does Crush compare to OpenCode?
Crush focuses on terminal aesthetics and MCP integration, while OpenCode offers more workflow automation out of the box. Both are open source with BYO-key support. Crush's TUI is significantly more polished.
What works
- Stunning terminal UI inherited from Charm ecosystem
- Free and open source with no usage restrictions
- BYO-key: full choice of models and providers
- MCP extensibility for custom tooling
- Active community and polished DX
What doesn't
- Steep learning curve for developers new to terminal TUIs
- Requires manual API key setup—no built-in model credits
- Smaller plugin ecosystem compared to mature CLI agents
The verdict
Crush is an excellent choice if you live in the terminal, prefer a beautiful TUI over web UIs, and want the freedom to choose any LLM backend. It's less suited for developers who want a turnkey managed service or need integrated IDE-level debugging. For terminal die-hards and Charm fans, Crush delivers the most glamorous AI coding experience on the command line.
FAQ
- Is Crush free to use?
- Yes. Crush is free and open source. You only pay for the API tokens consumed by the LLMs you configure.
- Can I use Crush without an API key?
- No. Crush itself contains no built-in models; you must provide your own API keys for a supported provider.
- How does Crush compare to OpenCode?
- Crush focuses on terminal aesthetics and MCP integration, while OpenCode offers more workflow automation out of the box. Both are open source with BYO-key support. Crush's TUI is significantly more polished.
Keep reading
- Kiro AIcodingJul 19, 2026
Kiro Review 2026: AWS's Spec-Driven AI IDE
Kiro brings a spec-first paradigm to AI coding that reduces guesswork. It excels where requirements are well-defined, but fal where speed and flexibility matter most. Worth trying for free if you value contracts over conversation.
4.2/ 5 - Augment CodecodingJul 19, 2026
Augment Code Review 2026: Context Engine for Big Codebases
Augment Code delivers on deep context retrieval for large monorepos, reducing manual context gathering. The free tier is great for testing. If you work on a large codebase, the paid plan provides clear value; otherwise, lighter tools remain more practical.
4.1/ 5 - Amazon Q DevelopercodingJul 18, 2026
Amazon Q Developer Review 2026: AWS's AI Coder Tested
Amazon Q Developer is the best AI coding assistant for AWS-centric teams, offering unparalleled context awareness and infrastructure generation. For general software development, its abilities are decent but not best-in-class. Use it as a complement to a general-purpose assistant if your stack is heavily AWS.
4.0/ 5 - TabninecodingJul 18, 2026
Tabnine Review 2026: Is the Privacy-First AI Assistant Still Relevant?
Tabnine remains the go-to AI coding assistant for organizations that prioritize data privacy over raw completion power. Its local and air-gapped deployment capabilities are unmatched in 2026, but its model quality—especially for multi-line suggestions—is not on par with cloud-based competitors like GitHub Copilot. Choose Tabnine if compliance requirements force you to keep code off external servers; otherwise, you may find Copilot or similar tools more productive.
3.8/ 5