Charm
Also known as: Charm, Charmbracelet, Charm (charmbracelet), Charmbracelet, Inc., Crush, charm.land, Charm Hyper, Catwalk, Fantasy
Terminal native AI coding agent from Charm with a broad choice of model providers, mid session model switching that keeps context, language server code intelligence, MCP and Agent Skills extensibility, and permission prompts before tool calls, running locally including air gapped.
Charm, the company behind the charmbracelet open source ecosystem, makes the command line glamorous with a suite of Go based terminal libraries and tools. Its flagship agentic product is Crush, a terminal native AI coding agent that connects to large language models and gives them tools to read, write and execute code directly in the terminal. Crush lets developers choose among Anthropic, OpenAI, Gemini, Bedrock, Copilot, Vertex, Groq and local models and switch mid session, with a Catwalk provider database that keeps model metadata current.
It pulls in code intelligence through Language Server Protocols, reads project context files, and persists sessions and history in SQLite. It extends through Model Context Protocol servers and agent skills, with two built in skills for its own configuration and hooks, and gives teams fine grained control over permissions and repo local configuration. Because it runs locally in the terminal it supports air gapped and local model workflows, and it works across macOS, Linux, Windows and the BSDs.
Crush began as a Go agent called OpenCode, built by Kujtim Hoxha, who brought it to Charm in July 2025, where it was renamed Crush. The OpenCode that Anomaly ships today is a separate codebase, and Crush can use OpenCode's Zen and Go model plans as one of its providers. Charm also runs Hyper, its own paid model service for Crush, with Shred plans from $10 to $200 a month, per seat Business plans with zero data retention, and prepaid credits from $5. Several terminal clients pointed at the same project share one workspace, with its sessions, message history and permission queue.
From our comparison video
Crush vs OpenCode: Language Servers or Background Subagents?
Our 2 minute video puts Charm side by side with OpenCode. It covers what each one does and how each is priced, then when to choose which.
Recorded in October 2026. The profile here is live and updates as research is verified.
Vendor details
Canonical URL
https://charm.land
Category
Coding agent
Subcategory
Terminal-native coding agent (BYOK, multi-provider)
Funding status
Seed stage, based in New York, founded by Christian Rocha; the round amount is not stated. Charm is the company behind a widely adopted open-source terminal tooling ecosystem including Bubble Tea, Lip Gloss, Glow and Gum, with the vendor stating the Charm ecosystem powers more than 25,000 applications. Crush itself has about 28,000 GitHub stars and 2,200 forks since the repository was created in May 2025. Charm now also operates Hyper, a subscription model provider with a free tier, which is its first documented commercial product line.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Crush connects to about 20 named LLM providers, including Charm Hyper, Anthropic, OpenAI, Gemini, Bedrock, Vertex, Azure OpenAI, Groq, OpenRouter and OpenCode Zen and Go, plus compatible endpoints and local models it discovers automatically. It uses Language Server Protocols for code intelligence, extends via Model Context Protocol servers over stdio, http, and sse, and can call local tools like git, docker, and npm.
In practice
You want a coding agent in your terminal that keeps your code on your own machine. Crush runs locally, works with local models through Ollama or LM Studio, and can run air gapped once provider auto update is switched off.
You want an AI review on every pull request without anyone starting it. Crush's non interactive crush run mode fits into CI, and a published GitHub Actions workflow runs it on each pull request to review the diff.
You want to move between Anthropic, OpenAI and a local model without losing your place. Crush lets you pick from about 20 providers or your own compatible endpoint, and switch models mid session while keeping the conversation.
Sources & related URLs
Related / legacy domains
Agentic Index coverage score
10.0 / 14 capabilities · 71%
| Integrations & Tool Calling | Full |
|---|---|
|
Crush connects to MCP servers over stdio, HTTP and SSE, uses language servers for diagnostics and references, fetches web pages, and runs the developer's own toolchain through its bash tool, alongside built in tools to view, grep, glob, edit and write files. SourceCrush docs (configuration/mcp, tools overview) and github.com/charmbracelet/crushread 2026-09-29 |
|
| Workflow Orchestration | Partial |
|
One agent works through a loop of file, search, language server and shell tools, with several sessions per project and models switchable mid session. There are no subagents, handoffs, branching or coordination across agents. SourceCrush docs (tools overview, sessions) and github.com/charmbracelet/crushread 2026-09-29 |
|
| Knowledge Grounding & RAG | Partial |
|
Context is assembled per run from the repository through view, grep and glob tools, language server diagnostics and references, and project context files such as AGENTS.md and CRUSH.md. There is no index, graph or embeddings layer that persists and can be queried. SourceCrush docs (context files, tools) and github.com/charmbracelet/crushread 2026-09-29 |
|
| Human Oversight & Guardrails | Full |
|
Crush asks the developer for permission before it runs tool calls, including in non interactive runs, with an allow list for trusted tools and a --yolo flag that skips the prompts only when the developer chooses it. Sourcegithub.com/charmbracelet/crush and Crush docs (permissions, crush run)read 2026-09-29 |
|
| Security, Identity & Governance | Partial |
|
Provider keys are supplied locally on the developer's own machine, where Crush runs. Usage metrics can be switched off, prompts and responses are never collected, and ignore files limit what Crush reads. On Charm's optional Hyper service, Business plans and prepaid credits use zero data retention, while Shred plans may be used to train Charm models from mid October 2026. Hyper Enterprise lists SAML single sign on and audit exports. No SOC 2, ISO or other attestation or trust center is published. Sourcegithub.com/charmbracelet/crush, hyper.charm.land and hyper.charm.land/prose/privacyread 2026-10-25 |
|
| Observability & Auditability | Full |
|
Crush writes a per project JSON log with timestamps and session IDs that records each tool call, such as a bash command, and each prompt sent and response received. The log can be viewed with crush logs, tailed or followed live, and debug mode adds language server traffic. Sessions and their history are kept in a local SQLite database. SourceCrush docs (logging, sessions) and github.com/charmbracelet/crushread 2026-09-29 |
|
| Memory & State Persistence | Partial |
|
Sessions persist across restarts in a per project SQLite database and can be reopened, and several clients pointed at the same project share one workspace, with its session list, message history and permission queue. The model remembers only the current session, and no context carries from one session to another. Project context files are instructions the developer writes. Sourcegithub.com/charmbracelet/crushread 2026-10-22 |
|
| Deployment & Data Residency | Full |
|
Code stays on the developer's machine because Crush runs locally in the terminal. It works air gapped once provider auto update is switched off and providers are updated manually, and can use local models through Ollama or LM Studio. Sourcegithub.com/charmbracelet/crush and Crush docs (air gapped, local models)read 2026-09-29 |
|
| Prebuilt Agents / Templates / Packs | Partial |
|
Skills follow the open Agent Skills standard, with SKILL.md packages discovered from a skills directory. Crush now bundles two built in skills for its own configuration and hooks (crush-config and crush-hook), but there is no catalog of skills or agents, so customers assemble the rest themselves. Sourcegithub.com/charmbracelet/crushread 2026-10-25 |
|
| Triggers & Channel Coverage | Full |
|
The crush run command is a non interactive mode for scripts and CI/CD pipelines. A published GitHub Actions workflow runs it on every pull request to review the diff, so work reaches the agent without a person starting it. The interactive terminal is the other surface. Sourcegithub.com/charmbracelet/crushread 2026-09-29 |
|
| Model Flexibility & Routing | Full |
|
The developer picks the provider and model from about 20 named providers, including Charm's own Hyper, Anthropic, OpenAI, Google Gemini, Amazon Bedrock, Azure OpenAI, Groq, OpenRouter, Hugging Face and OpenCode Zen and Go, or uses custom OpenAI compatible or Anthropic compatible endpoints. Local models run through llama.cpp, LM Studio, LiteLLM or Ollama, which Crush discovers automatically. Models can be switched mid session while keeping the conversation. Sourcegithub.com/charmbracelet/crushread 2026-10-22 |
|
| APIs / SDKs / MCP Extensibility | Full |
|
Fantasy, the Go agent framework Charm built to power Crush, is a library with a multi provider API and custom tools. Crush can be driven from scripts through crush run with stdin and stdout, and extended with Agent Skills. MCP support lets Crush reach other systems. Sourcegithub.com/charmbracelet/fantasy and Crush docs (cli/run)read 2026-09-29 |
|
| Testing, Debugging & Optimization | Partial |
|
The agent runs project commands and tests through its bash tool and reads language server diagnostics, and debug logging exposes agent and language server behavior. These check the customer's code and help troubleshoot a run, but there is no harness to evaluate or regression test the agent's own behavior. Sourcegithub.com/charmbracelet/crushread 2026-09-29 |
|
| Browser / Computer-use | Not documented |
|
Everything runs through file tools, shell commands, language servers, MCP servers and a fetch tool that retrieves pages. Crush does not control a browser or operate software through a graphical interface. Sourcegithub.com/charmbracelet/crushread 2026-09-29 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
Background subagents arrive in Crush v0.100.0: the agent can dispatch detached subagents that keep working while the user checks on them, sends them follow up messages mid run, stops them or moves them to another model and provider. Each subagent has its own name, its progress shows in the transcript, sidebar and header, and its report returns to the parent agent. The release also adds Sysadmin mode, a permission mode in which Crush can run any command on the system.
Bears on: Workflow orchestration
View sourceCrush v0.99.0 widens hooks so teams can run their own checks on session, turn and permission events, and validates hook event names against a known list. The release adds Catwalk v3 support, where custom model providers are defined by the API format they speak rather than by vendor name, and existing configs migrate on their own. An app that embeds Crush can also answer the agent's permission prompts itself.
Bears on: Human approval / guardrails
View sourceCharm released version 0.88.0 of the Crush terminal coding agent, introducing a scriptable configuration system via crush.sh and crushrc. The update adds explicit tool permissions, allowing users to apply "deny" rules to completely hide specific tools from the AI agent. The release also includes verb-first setup commands for managing MCP servers and LSPs, as well as auto-retry logic for AWS SSO logins.
Bears on: Security / enterprise
View sourcePricing
Crush is free and source available with the developer's own model keys. Charm's optional Hyper inference plans start at $10 a month.
Crush itself is free. Model usage is paid to the chosen provider, or to Charm through Hyper: monthly Shred plans with Hypercredits, per seat Business plans, or prepaid credit bundles.
Cost watchouts
The tool is free, but running Crush means paying for model usage, through the developer's own provider or Hyper credits, which scales with how much coding work you drive through it.
Variable cost rationale
Crush meters nothing itself. Developers bring their own model keys, so the running cost is LLM token usage paid to the chosen provider, which scales directly with coding activity. The provider's own terms are the only limit on that spend. On Charm's Hyper service, Shred plans carry a Hypercredit allowance (Shred Classic gives 250 a week with a $7.50 rolling five hour cap), one Hypercredit is 5 cents, and prepaid bundles from $5 set a ceiling on what is spent.
Sales call required
No, self serve available
Free / trial
Crush is free and source available with your own model keys. Hyper no longer has a free monthly credit allowance.
Lowest paid plan
Hyper Shred Diet at $10 a month (180 Hypercredits), or prepaid credit bundles from $5
Key ambiguities
Charm publishes no commercial price for Crush itself. The paid offers are Hyper inference plans and credits. Shred plans may be used to train Charm models from mid October 2026, while Business plans and prepaid credits stay zero data retention.
Related vendors
- Cognition — Maker of Devin, an autonomous AI software engineer
- 10Web — Agentic website platform whose AI agents build, host and manage…
- AgentUI — Managed AI app builder for operations teams that generates and hosts…
- Aider — Open source, model agnostic terminal coding agent that edits your…
- Amp — Amp is a coding agent from Amp Frontier Corporation that runs each…
- Anthropic Claude Code — Anthropic's agentic coding system across terminal, desktop, IDE, web…
Alternatives to Charm
The closest documented capability profiles to Charm among coding agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Poolside11.0 / 14Fuller documented coverage on Workflow Orchestration and Security, Identity & Governance
- Ellipsis10.5 / 14Fuller documented coverage on Workflow Orchestration and Security, Identity & Governance
- OpenCode10.5 / 14Fuller documented coverage on Workflow Orchestration and Security, Identity & GovernanceCharm vs OpenCode →Video
- Tabnine11.5 / 14Fuller documented coverage on Workflow Orchestration and Knowledge Grounding & RAG
- Augment Code12.0 / 14Fuller documented coverage on Workflow Orchestration and Knowledge Grounding & RAG
- Cline12.0 / 14Adds documented Browser & Computer Use
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded