CrushvsWarp
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
Crush
Charm · Terminal agent
#25OSSMCP
Panel
6.51 spec wins
- Reliability
- 6.0
- Usefulness
- 6.5
- Cost
- 7.3
- Longevity
- 6.2
“Has a --yolo flag, auto-discovers your local llama.cpp, and a license with MIT in the name that is not MIT yet.”
Warp
Warp · Terminal agent
#20OSSMCP
Panel
6.54 spec wins
- Reliability
- 6.5
- Usefulness
- 6.7
- Cost
- 5.5
- Longevity
- 7.2
“Scored 75.8 on SWE-bench Verified with a best-of-k wrapper, which is also how most of us get through code review.”
Spec by spec
| Spec | Crush | Warp |
|---|---|---|
| Architecture | ||
| Category | Terminal agent | Terminal agent |
| Runs | local | local, cloud |
| Platforms | macos, linux, windows | macos, linux, windows |
| Context window | not documented | not documented |
| Protocols | ||
| MCP client | Yes | Yes |
| MCP server | No | No |
| Capabilities | ||
| Runs terminal commands | Yes | Yes |
| Multi-file edits | Yes | Yes |
| Git operations | Yes | Yes |
| Browser control | NoThe built-in tools stop at web_fetch, web_search and download; there is no browser, CDP or Playwright tool. | YesComputer Use environments bundle Chromium plus the Playwright CLI and can attach over CDP, so the agent navigates, clicks, fills forms and reads pages. |
| Sandboxed execution | NoShell and edit calls are guarded by permission prompts (or skipped with --yolo), with no container or OS-level isolation. | YesCloud runs execute in a containerized sandbox built from a Docker image with no access to the local machine; local interactive sessions are not sandboxed. |
| Multi-agent orchestration | No | Yes |
| Headless / CI mode | Yes`crush run` executes a single non-interactive prompt, reads stdin and supports --quiet for scripts. | Yes |
| Models | ||
| Backbone | Claude, GPT, Gemini, any OpenAI-compatible or Anthropic-compatible provider | GPT, Claude, Gemini, Grok, open-weight models via Fireworks |
| Bring your own model | Yes | YesAny OpenAI-compatible endpoint (OpenRouter, LiteLLM, an internal gateway) plus BYO Anthropic, OpenAI or Google keys and custom routers. |
| Local models | YesOllama, LM Studio, llama.cpp and MLX providers are auto-discovered, and any OpenAI- or Anthropic-compatible base_url can be added. | YesOllama, LM Studio, vLLM and llama.cpp are supported through the custom inference endpoint, but only when exposed at a public URL via a tunnel, since localhost and private addresses are rejected. |
| Cost | ||
| Pricing model | byok | mixed |
| Starts at | $0/mo | $20/mo |
| Free tier | Yes | Yes |
| Bring your own key | Yes | Yes |
| Openness | ||
| Open source | Yes | Yes |
| License | FSL-1.1-MIT | AGPL-3.0 |
| GitHub stars | 28,444 | 65,338 |
Which one would each critic pick
| Critic | Crush | Warp | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 7.3 | 7.3 | no preference |
| El Crítico | 6.3 | 6.0 | Crush — A --yolo flag on an agent with no sandbox means a headless run is your shell with nobody watching; the LSP context is the reason to read past it. |
| El Profesor | 6.5 | 6.0 | Crush — Crush's use of language servers as the context source is the most principled retrieval choice among terminal agents, and its autonomy flag is the most accurately named. |
| La Inversora | 5.8 | 6.8 | Warp — Warp spent years building a terminal, open-sourced it under AGPL, and is now selling credits and $200 Max plans on top of it; the terminal is the distribution, the agent is the business. |
| La Jefa | 5.8 | 6.0 | Warp — SAML SSO and bring-your-own-key arrive at Business, $50 a seat with a 25-seat cap, so sixty engineers means Enterprise and a custom quote; approved with conditions. |
| El Hacker | 7.5 | 6.8 | Crush — Reads my LSPs, finds my llama.cpp server on its own, MCP in a JSON file, a --yolo flag, and a license that keeps me from forking it today; close to great, grudgingly. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.