Aquesta és una versió anterior. La versió actual és la 3000.11.228. Devin CLI 3000.11.228 · Data de publicació:
Fitxers d’instal·lació 2026.5.6-1
| Arquitectura | Tipus | Àmbit | Mida del fitxer | Baixa | SHA256 |
|---|---|---|---|---|---|
| 64 bits | ZIP → Portable | — | — | static.devin.ai | VirusTotal |
| ARM64 | ZIP → Portable | — | — | static.devin.ai | VirusTotal |
L’enllaç i l’SHA256 provenen del manifest comunitari del Gestor de paquets del Windows. Convé comparar el hash del fitxer baixat abans d’executar-lo.
Notes de la versió
Web search
Search the web directly from your Devin CLI sessions. The agent can look up documentation, find solutions, and pull in relevant information from the internet without leaving the terminal.
Added
- Built-in OAuth device flow for GitHub MCP server. devin mcp add github --url https://api.githubcopilot.com/mcp/ now authenticates via device flow (enter a code at github.com/login/device) without needing --oauth-client-id.
- /copy command to copy the last agent response to the system clipboard. Works over SSH connections and on Linux desktops.
- Numbered options in select prompts can now be picked directly with the 1-9 keys instead of arrowing + Enter. The shortcut is shown as a digit prefix on each option in non-search prompts.
- web_search tool for searching the web during agent sessions.
Fixed
- Cancelling a session now also stops running subagents instead of letting them continue in the background
- Shell commands that redirect output to /dev/null (e.g. 2>/dev/null, >/dev/null, &>/dev/null) no longer prompt for write permission to /dev/null.
- Edit tool previews now show correct file line numbers instead of always starting from 1.
- Output token limit raised from 16k to match each model’s actual capacity (128k for Opus, 64k for Sonnet), preventing premature response truncation.
- Option+Backspace now correctly deletes words in select menus (user question “Other” field and search) on BS-mode terminals, instead of inserting ‘h’.
- Slash command output now has consistent visual separation from the prompt, matching how agent responses are displayed.