Rating · evidence to 26 Sept 2026
Not enough evidence
- Trust B
The Register / Coding agents / Crush
Register entry · Coding agents · updated 26 Sept 2026
Crush
Charm (Charmbracelet, Inc.) · Coding agents
Crush is a Go terminal coding agent built on Charm's TUI libraries. It connects to Anthropic, OpenAI, Gemini, OpenRouter, Bedrock, Azure, local models and other OpenAI/Anthropic-compatible endpoints, and supports LSP context, MCP servers and Agent Skills. It asks permission before tool calls by default (a --yolo flag skips prompts). The code is source-available under FSL-1.1-MIT; Charm sells optional inference through its Hyper service.
What reviewers and users praise
- Polished terminal UI (Micah Walter, aicoolies)
- Works well with open-weight models such as GLM (HN users)
- Model flexibility with BYOK (aicoolies, Ry Walker)
What they criticise
- Not OSI open source (FSL-1.1-MIT); criticised on HN
- Lags on features such as subagents/slash commands per some users (HN)
- Token use and startup speed complaints on GitHub
Experts vs the crowdneeds both scores
Experts—
Crowd—
| Price | Free, paid from $5 |
|---|---|
| Use it via | CLI |
| Runs on | macos, linux, windows |
| Licence | FSL-1.1-MIT |
| Updated | Evidence collected 26 Sept 2026 |
How the rating was worked out
Weights for Coding agents: adoption 35%, experts 40%, crowd 25%. Fewer than two pillars have enough evidence. The method.
| Adoption & momentumweight 35% | 64 | GitHub stars 28k → 74 · npm weekly downloads 6.2k → 22 · 22 releases in 90 days → 100reach × 0.7 + momentum × 0.3 |
|---|---|---|
| Expertsindependent reviews · weight 40% | — | Only 2 independent reviews (3 needed) |
| Crowdcommunity sample · weight 25% | — | Community sample from a single platform |
| Rating | — | Not rated: fewer than two pillars have enough evidence. |
Evidence checks. 1 review from tool directories, AI-written pages or unnamed authors not counted. 2 GitHub issues not counted as sentiment. How evidence is checked.
Trust & safety
Seven dimensions graded A–D from documented facts. Overall: B.
B
Permission model
README: Crush asks permission before executing tool calls by default; --yolo skips all prompts. No default sandbox documented. source
B
Data access scope
Runs as the user in the working directory with broad filesystem/shell access; behaviour documented in README. source
A
Data storage
Runs locally; sessions stored locally; no required cloud storage (BYOK). source
n/e
Data retention & training
Crush metrics are pseudonymous usage metadata only; prompts and responses never collected (README). Model-side retention depends on chosen provider; Charm Hyper advertises zero data retention. No explicit no-training statement verified. source
A
Incident history (24 months)
No GitHub security advisories published on charmbracelet/crush as of 2026-09-26. NVD not searched separately. source
D
Compliance
No SOC 2 / ISO 27001 attestation found for Crush; Hyper claims GDPR compliance only. source
B
Transparency
Source-available under FSL-1.1-MIT (not an OSI licence) with detailed public GitHub release notes. source
The experts
2 named independent reviews found; dated reviews from the last 12 months are scored.
- positive
It's the terminal AI assistant I wanted to build, executed with the level of polish and thoughtfulness that Charm brings to everything they do.
Micah Walter · 17 Feb 2026 · micahwalter.com - positive
Expect Crush to keep gaining on the strength of Charm's craft and cadence.
Ry Walker Research · 11 Jun 2026 · rywalker.com
The crowd
10 coded posts from hackernews.
6 positive3 mixed1 negative
- positive
I've been using charm's Crush with GLM for several months and it's been working great.
Hacker News comment · 12 Jun 2026 · news.ycombinator.com - positive
My experience using GLM-4.6 with Charm Crush has been absolutely incredible, especially with high thinking.
Hacker News comment · 16 Oct 2025 · news.ycombinator.com - negative
Crush is notably and disappointingly proprietary, being a fork (by the original author, who was hired by Charm) of the open-source OpenCode
Hacker News comment · 12 Nov 2025 · news.ycombinator.com - mixed
They're still lacking slash commands, sub agents etc (since they don't own their own model), but they do integrate language servers
Hacker News comment · 29 Oct 2025 · news.ycombinator.com - mixed
My favourite agent crush[0] has lsp support for a while. I’ve not noticed the agent deciding to use it all that much.
Hacker News comment · 22 Dec 2025 · news.ycombinator.com - positive
I use crush as coding agent but removed the blocked commands because I also do a lot of system administration. Love it.
Hacker News comment · 15 Jun 2026 · news.ycombinator.com - positive
Check out the Crush agent harness. It's very impressive.
Hacker News comment · 3 May 2026 · news.ycombinator.com - mixed
They work well, except that the agent will often simply call wait() immediately after
Hacker News comment · 23 Sept 2026 · news.ycombinator.com
Pricing
As published, checked 26 Sept 2026.
| Plan | Price | Notes |
|---|---|---|
| Crush (BYOK) | Free | Free to use with your own provider keys. |
| Hyper Free | Free | 100 Hypercredits per month (1 Hypercredit = 5 cents). |
| Hyper Subscription | $20 / monthly | 250 Hypercredits, refreshing daily. |
| Hyper prepaid bundles | $5 / one-time | $5/100, $10/200, $20/400 Hypercredits; never expire. |
Recent changes
Ranked alongside
Other ranked entries in Coding agents.
OpenCodeAnomaly75
Claude CodeAnthropic73
Zed (Agent Panel)Zed Industries73
GitHub CopilotGitHub (Microsoft)72
Codex (CLI and app)OpenAI71Sources
Every figure above traces to one of these, observed up to 26 Sept 2026.