Crush
Crush is an agentic coding assistant that runs in your terminal with live model switching and LSP-aware context.
For terminal-first developers who want an agent sitting next to git, your shell and your language servers, Crush is the pick, and mid-session model switching earns its place. The cost is on you: provider keys, per-token spend, no admin console. Need GUI polish or team seat billing? Cursor or Copilot.
Verified 6h ago · liveness 85/100 · cite: rightaichoice.com/tools/crush
- Terminal-first developers who want agentic coding in the CLI rather than an IDE
- Engineers who swap between LLMs mid-task and need session context preserved
- Developers working headless or over SSH who want assistance where they already are
- NixOS and Home Manager users wanting declarative setup via the Charm NUR module
- Anyone who wants a graphical, IDE-style assistant and avoids the terminal
- Beginners after a zero-config, sign-up-and-go cloud coding tool
- Teams needing centralised billing, admin dashboards or usage governance
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Crush if you want a GUI assistant with centralised seat billing and no API keys to manage — it's built for people who work in a terminal and are happy to pay model providers per token.
The MIT client is free, but every request is billed by your model provider — a heavy refactor session on a frontier model can cost more than an IDE assistant's flat seat fee.
The Crush client itself costs $0 under MIT, so it's the cheapest entry point among agentic coding assistants — but the real cost is your provider bill. Developers who want predictable flat-rate billing are better served by GitHub Copilot or Cursor seat pricing; high-volume users on their own API keys may come out ahead, and Charm's Hyper subscription adds a managed provider option with a free tier.
In short
Crush — Crush is an agentic coding assistant that runs in your terminal with live model switching and LSP-aware context. Best for Terminal-first developers who want agentic coding in the CLI rather than an IDE, Engineers who swap between LLMs mid-task and need session context preserved, Developers working headless or over SSH who want assistance where they already are. Free to use.
No head-to-head comparison for Crush yet — the alternatives below are the closest matches we list.
What's new in Crush
Checked 2 days agoAcross the latest 2 updates: 2 changelog entries.
Crush README documents .agents folder and shell command approval gates
The Crush README now documents .agents folders for defining custom agents for specialised tasks alongside shell command approval gates that keep execution under control.
Official NixOS and Home Manager modules available via the Charm NUR
Crush ships NixOS and Home Manager modules through nur.repos.charmbracelet.crush, with the module auto-detecting home manager versus NixOS context so the same import works either way.
What people actually say about Crush — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
22 mentions across 6 sources (Hacker News, YouTube, Product Hunt, App Store, Stack Overflow, Lemmy), 76 more we could not attribute · researched Oct 14, 2026.
Weighted by the 98 posts each of 6 sources contributed.
- +MIT-licensed and free to use — you pay only for the LLM API calls you make
- +Multi-model switching mid-session preserves context, enabling real side-by-side comparison
- +LSP-enhanced context ships with example configs for gopls (Go) and nil (Nix)
- +MCP support over http, stdio and sse matches the current agent-tooling standard
- +Runs on an unusually wide OS spread — macOS, Linux, Windows, BSDs, Android, NixOS
- −The name 'Crush' collides with Candy Crush, a Cdrama, and a job tool — search is hopeless
- −Scraped data contains virtually zero genuine user reviews of this coding agent
- −Bring-your-own-key model means LLM costs are entirely on you and can surprise you
- −Terminal-only UX means no visual diff review like Cursor or Copilot offer
- −Requires you to configure providers, env vars and LSP servers yourself
- • LLM API usage charges from Anthropic, OpenAI, Google or OpenRouter — uncapped and usage-dependent
- • Time cost of configuring providers, LSP servers, MCP endpoints and custom agents
- • Potential future costs if Hyper's free tier narrows — the client is free but the provider is the vendor's monetization path
Viability Score
How well maintained and how widely used is Crush? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Agentic coding assistant that runs in your terminal
- Multi-model support: Anthropic, OpenAI, Google Gemini, OpenRouter and more
- Add custom providers via OpenAI- or Anthropic-compatible APIs
- Switch LLMs mid-session while preserving conversation context
- Model picker for selecting and authenticating providers
- Session-based workflows for multiple contexts per project
- LSP-enhanced context with example configs for gopls (Go) and nil (Nix)
- Extend capabilities with MCP servers over http, stdio and sse
- Define custom agents for specialised tasks in .agents folders
- Shell command approval gates to keep agent execution supervised
- NixOS and Home Manager modules via the Charm NUR with auto-detected context
- Runs on macOS, Linux, Windows (PowerShell and WSL), Android, FreeBSD, OpenBSD and NetBSD
- Install via Homebrew, npm, Arch, Nix, FreeBSD pkg, Winget, Scoop, Debian, Fedora or go install
- TUI with compact mode and debug options
- Hyper provider with free tier and zero data retention
About Crush
Crush puts an agentic coding assistant inside your terminal instead of an IDE sidebar. Charmbracelet built it on the same Charm TUI stack that powers thousands of CLI apps, and the result is a shell-native agent: it reads your code, queries your language servers for symbol context, and runs shell commands behind approval gates you control. It runs first-class on macOS, Linux, Windows (PowerShell and WSL), Android, FreeBSD, OpenBSD and NetBSD. Model choice drives the design. You can point Crush at Anthropic, OpenAI, Gemini, OpenRouter or any custom endpoint speaking an OpenAI- or Anthropic-compatible API, and swap LLMs mid-session without losing conversation context. That mid-task comparison is the feature most terminal coders actually use day to day. It leans on tooling you already have. LSP integration adds structure and symbol context (the README ships example configs for gopls in Go and nil in Nix), MCP servers over http, stdio and sse extend reach, and .agents folders let you define custom agents for specialised jobs. Sessions keep separate project contexts, and shell approval gates keep execution supervised. NixOS and Home Manager modules via the Charm NUR let you configure the whole thing declaratively, with the module auto-detecting which context it's in. Installation is wide: Homebrew, npm, Arch, Nix, Winget, Scoop, Debian, Fedora, FreeBSD pkg or go install, plus binaries for illumos and Solaris. Charm's own Hyper provider is subscription-based with a free tier and zero data retention. Against Cursor or GitHub Copilot, Crush trades GUI polish and centralised seat billing for openness and direct control over which model touches your code.
Behind the Verdict
Where Crush earns its keep is the moment you're already three commands deep in a shell and don't want to context-switch to an editor. It reads the repo, pulls symbol context from your LSP, and proposes commands you approve before they run — no IDE window required. The mid-session model switch is the part I'd actually miss. Midway through a refactor you can compare how two models handle the same problem without losing the thread. In practice that changes how you debug: you stop guessing which model to commit to up front. Pick Crush if you live in a terminal or work headless over SSH, if you use NixOS or Home Manager and want declarative config through the Charm NUR module, or if you want control over which model sees your code. The .agents folders plus MCP servers give you room to shape it into something bespoke. Pass if you need a graphical assistant, want zero-config sign-up-and-go, or manage a team that needs centralised billing and usage governance. Crush assumes you'll bring your own provider keys and watch per-token spend yourself. The closest alternative is Cursor: same ambition, different centre of gravity. Cursor gives you a polished editor and seat management; Crush gives you openness and direct model control at the cost of setup work. Claude Code is the nearer rival for terminal agentic coding — worth comparing on provider flexibility and LSP depth before you commit. Real-world caveats: expect to configure providers, LSP commands and MCP servers before it feels natural. The approval gate is a safety net, not a substitute for reading what the agent wants to run. And on exotic platforms like illumos, native OS notifications are unavailable though terminal notifications still fire.
Researching Crush? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Crush actually fits — and what changes day-one when you adopt it.
SSH into the box, install Crush via Homebrew or a Debian package, set the OpenAI or Anthropic API key as an environment variable, then ask it to trace why a service is returning 500s while LSP-fed context pulls in real symbol definitions.
Outcome: Diagnosis happens where the code runs, with no IDE extension to install on a headless machine and no file syncing between local and remote.
Run Crush in one tmux pane next to the editor and a test runner, using the gopls LSP config so Crush sees the same symbol graph, and approve shell commands before they execute.
Outcome: Agentic edits, test runs and shell commands stay in the terminal layout they already use, with approval gates keeping execution under control.
Import the Charm NUR Home Manager module, declare programs.crush with provider settings, LSP entries for gopls and nil, context_paths and compact_mode in configuration.nix.
Outcome: Crush is provisioned like the rest of the system — reproducible across machines, no manual dotfile wrangling.
Use Cases
- Run an AI coding assistant over SSH on a remote dev server without an IDE.
- Work inside tmux panes alongside your editor and shell, all in the terminal.
- Swap models mid-session while preserving context to compare output on the same task.
- Use LSP integration to feed Crush the same symbol and structure context your editor sees.
- Execute shell commands via the assistant with approval gates for safety.
- Define custom agents for specialised tasks using .agents folders.
- Maintain multiple session contexts for different features or bug fixes.
- Extend Crush with MCP servers over http, stdio or sse for external tools and data.
Models Under the Hood
as of 2026-10-16
Limitations
- A TUI has inherent limits: reviewing a 500-line diff or a visual design spec is awkward compared to an IDE.
- Crush is a younger project than Aider or Claude Code, so expect rough edges on the long tail of commands.
- Shell execution is gated behind approval, which is safer but requires you to configure those gates deliberately.
- There is no centralised team management or usage dashboard, so cost visibility across a team is on you.
- The MIT client is free, but you supply API keys and pay providers per token unless you use Hyper's subscription.
as of 2026-10-01
Verification history
We have re-verified Crush 22 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 22 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Crush tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Open Source
$0
Ideal for
Individual developers and small teams who already have Anthropic, OpenAI, Gemini or OpenRouter API keys, or who run local models, and want no monthly seat fee.
What this tier adds
Starting tier: the MIT-licensed client is $0 — you pay model providers directly per token.
Hyper
Subscription with free tier
Ideal for
Crush users who'd rather not manage their own provider keys and want a managed option with zero data retention, starting on a free tier.
What this tier adds
Adds Charm's official Crush provider on a subscription with a free tier, ZDR and a GDPR-compliant design, so you skip buying and rotating your own API keys.
Where the pricing makes sense
The company stage and team size where Crush's pricing actually pencils out — and where peers do it cheaper.
The Crush client itself costs $0 under MIT, so it's the cheapest entry point among agentic coding assistants — but the real cost is your provider bill. Developers who want predictable flat-rate billing are better served by GitHub Copilot or Cursor seat pricing; high-volume users on their own API keys may come out ahead, and Charm's Hyper subscription adds a managed provider option with a free tier.
Setup time & first value
How long it actually takes to get something useful out of Crush — broken out by persona, not the marketing-page minute.
Terminal-first developers with an API key on hand: about 5–10 minutes — install via Homebrew, npm or your distro package, export a provider key, and you're in the TUI. NixOS and Home Manager users: roughly 15–30 minutes to import the Charm NUR module and declare programs.crush in configuration.nix. Teams evaluating centralised governance: not applicable — there's no admin console to set up.
Switching to or from Crush
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From Aider: install Crush alongside it, point the same provider API key at Crush's provider config, and move recurring prompts into .agents folders.
- →From GitHub Copilot in the terminal: keep your Copilot seat if you like, install Crush via Homebrew or npm, and set an OpenAI or Anthropic key for the model you want Crush to use.
- →From Cursor: keep the IDE for visual diffs and add Crush in tmux for SSH and headless work where Cursor can't run.
- →From a Nix flake setup: swap your ad-hoc derivation for the official nur.repos.charmbracelet.crush module so upgrades track the NUR.
- ↗To Aider: port your .agents prompts into Aider's conventions and reuse the same provider API keys; Aider is the more mature project if you hit rough edges in Crush.
- ↗To Claude Code: keep your Anthropic key and move terminal workflows over if you want a vendor-managed agent with a longer track record.
- ↗To Cursor or GitHub Copilot: give up bring-your-own-model flexibility in exchange for a GUI, diff review and flat per-seat billing.
Integrations
Resources & Guides
- Resourcegithub.com
GitHub - charmbracelet/crush: Glamourous agentic coding for all 💘
Glamourous agentic coding for all 💘. Contribute to charmbracelet/crush development by creating an account on GitHub.
- Resourcegithub.com
crush/README.md at main · charmbracelet/crush
Glamourous agentic coding for all 💘. Contribute to charmbracelet/crush development by creating an account on GitHub.
Tutorials & Learning
YouTube returned 6 videos for “Crush”, and we withheld 6: 6 could not be judged, because “Crush” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Crush.
Official links
Tools that pair well with Crush
Common stack mates teams adopt alongside Crush, with the specific reason each pairing earns its keep.
Bito
Bito Governor is an AI model router with a code context engine that cuts coding-agent spend by 40-70%.
Poolside AI
Poolside builds open-weight agentic coding models — Laguna XS 2.1 (33B) and Laguna S 2.1 (118B) — for code that stays inside your security
Claude Code
Claude Code is Anthropic's agentic coding assistant for multi-file repo work from the terminal, IDE, browser, or cloud.
Alternatives to Crush
View all
Bito
Bito Governor is an AI model router with a code context engine that cuts coding-agent spend by 40-70%.
Poolside AI
Poolside builds open-weight agentic coding models — Laguna XS 2.1 (33B) and Laguna S 2.1 (118B) — for code that stays inside your security
Claude Code
Claude Code is Anthropic's agentic coding assistant for multi-file repo work from the terminal, IDE, browser, or cloud.
Frequently Asked Questions
Best-of guides
Used Crush? Help shape our editorial sentiment research.