You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
For those interested mistral-vibe (installed with uv pip install) works out-of-the-box, even under OpenBSD, and can be configured to use local (ollama) or remote (openai-compatible) models
I think this is a great idea. I've actually been thinking about developing it myself, because CLI environments (like on servers) are where this is really needed.
If the team could provide some guidance on the architectural approach or how to get started, I'd be happy to contribute and help build it.
Most of the other "AI Coding Agents" (Cline, Continue, KiloCode, Claude, Codex, etc.) have a CLI version.
That very useful for those that do not use VSCode.
Would be nice to be able to run Zoo-Code outside of VSCode.
Hey @CipherChaos this would be a great addition. What would be needed is a deep understanding on the vscode integration points of zoo code, and we would need to refactor into a common codebase that can be used by both the plugin and cli, so the core logic is shared between them. We would need to do this without breaking exising Zoo functionality. Please join our discord and we can collaborate there
I also have a need for this and so worked with my LLM-friend to refactor the existing CLI (apps/cli) to support my use case (AWS Bedrock). It builds fine and pass tests; I've been using it for a few days now to replicate the /loop feature of claude code. @navedmerchant what is the proper etiquette for offering y'all this code? Should I fork and submit a PR? Clone and make a feature branch? Happy to contribute however you're comfortable with.
`❯ node apps/cli/dist/index.js --help
Usage: roo [options] [command] [prompt]
Roo Code CLI - starts an interactive session by default, use -p/--print for non-interactive output
Arguments:
prompt Your prompt
Options:
-V, --version output the version number
--prompt-file <path> Read prompt from a file instead of command line argument
--create-with-session-id <session-id> Create a new task with a specific session ID (must be a UUID)
--session-id <session-id> Resume a specific task by session ID
-c, --continue Resume the most recent task in the current workspace (default: false)
-w, --workspace <path> Workspace directory path (defaults to current working directory)
-p, --print Print response and exit (non-interactive mode) (default: false)
--stdin-prompt-stream Read NDJSON commands from stdin (requires --print and --output-format stream-json) (default: false)
--signal-only-exit Do not exit from normal completion/errors; only terminate on SIGINT/SIGTERM (intended for stdin stream harnesses) (default: false)
-e, --extension <path> Path to the extension bundle directory
-d, --debug Enable debug output (includes detailed debug information) (default: false)
-a, --require-approval Require manual approval for actions (default: false)
-k, --api-key <key> API key for the LLM provider
--provider <provider> API provider (anthropic, openai-native, gemini, openrouter, etc.)
-m, --model <model> Model to use (default: "anthropic/claude-opus-4.6")
--mode <mode> Mode to start in (code, architect, ask, debug, etc.) (default: "code")
--terminal-shell <path> Absolute path to shell executable for inline terminal commands
-r, --reasoning-effort <effort> Reasoning effort level (unspecified, disabled, none, minimal, low, medium, high, xhigh) (default: "medium")
--consecutive-mistake-limit <limit> Consecutive error/repetition limit before guidance prompt (0 disables the limit)
--exit-on-error Exit on API request errors instead of retrying (default: false)
--ephemeral Run without persisting state (uses temporary storage) (default: false)
--oneshot Exit upon task completion (default: false)
--output-format <format> Output format (only works with --print): "text" (default), "json" (single result), or "stream-json" (realtime streaming) (default: "text")
-h, --help display help for command
Commands:
list List commands, modes, models, or sessions
upgrade Upgrade Roo Code CLI to the latest version
auth Inspect or remove legacy Roo auth tokens`
`❯ node apps/cli/dist/index.js --help
Usage: roo [options] [command] [prompt]
Roo Code CLI - starts an interactive session by default, use -p/--print for non-interactive output
Arguments:
prompt Your prompt
Options:
-V, --version output the version number
--prompt-file <path> Read prompt from a file instead of command line argument
--create-with-session-id <session-id> Create a new task with a specific session ID (must be a UUID)
--session-id <session-id> Resume a specific task by session ID
-c, --continue Resume the most recent task in the current workspace (default: false)
-w, --workspace <path> Workspace directory path (defaults to current working directory)
-p, --print Print response and exit (non-interactive mode) (default: false)
--stdin-prompt-stream Read NDJSON commands from stdin (requires --print and --output-format stream-json) (default: false)
--signal-only-exit Do not exit from normal completion/errors; only terminate on SIGINT/SIGTERM (intended for stdin stream harnesses) (default: false)
-e, --extension <path> Path to the extension bundle directory
-d, --debug Enable debug output (includes detailed debug information) (default: false)
-a, --require-approval Require manual approval for actions (default: false)
-k, --api-key <key> API key for the LLM provider
--provider <provider> API provider (anthropic, openai-native, gemini, openrouter, etc.)
-m, --model <model> Model to use (default: "anthropic/claude-opus-4.6")
--mode <mode> Mode to start in (code, architect, ask, debug, etc.) (default: "code")
--terminal-shell <path> Absolute path to shell executable for inline terminal commands
-r, --reasoning-effort <effort> Reasoning effort level (unspecified, disabled, none, minimal, low, medium, high, xhigh) (default: "medium")
--consecutive-mistake-limit <limit> Consecutive error/repetition limit before guidance prompt (0 disables the limit)
--exit-on-error Exit on API request errors instead of retrying (default: false)
--ephemeral Run without persisting state (uses temporary storage) (default: false)
--oneshot Exit upon task completion (default: false)
--output-format <format> Output format (only works with --print): "text" (default), "json" (single result), or "stream-json" (realtime streaming) (default: "text")
-h, --help display help for command
Commands:
list List commands, modes, models, or sessions
upgrade Upgrade Roo Code CLI to the latest version
auth Inspect or remove legacy Roo auth tokens`
@Nicramus I have a fork of this that I got working, need to find time to write up a proper PR. Hoping to get that done this week.
Sorry for the delay! Finally found some time to get this knocked out. - Rowena
Bedrock provider support for the CLI is now up as #1015.
It adds bedrock as a supported CLI provider with the full AWS auth stack (bearer token / AWS_BEDROCK_API_KEY, AWS_PROFILE, direct access keys, and the AWS SDK default credential chain), plus dynamic --provider help text and cross-region inference auto-detection for us./eu./apac. model IDs.
(Content added by Kenzie, LLM-based artificial assistant, on behalf of Rowena Day.)
Activity
liar666 commentedon Jun 29, 2026
For those interested
mistral-vibe(installed withuv pip install) works out-of-the-box, even under OpenBSD, and can be configured to use local (ollama) or remote (openai-compatible) modelsCipherChaos commentedon Jun 30, 2026
I think this is a great idea. I've actually been thinking about developing it myself, because CLI environments (like on servers) are where this is really needed.
If the team could provide some guidance on the architectural approach or how to get started, I'd be happy to contribute and help build it.
navedmerchant commentedon Jul 1, 2026
Hey @CipherChaos this would be a great addition. What would be needed is a deep understanding on the vscode integration points of zoo code, and we would need to refactor into a common codebase that can be used by both the plugin and cli, so the core logic is shared between them. We would need to do this without breaking exising Zoo functionality. Please join our discord and we can collaborate there
rutgoff commentedon Jul 6, 2026
I also have a need for this and so worked with my LLM-friend to refactor the existing CLI (apps/cli) to support my use case (AWS Bedrock). It builds fine and pass tests; I've been using it for a few days now to replicate the /loop feature of claude code. @navedmerchant what is the proper etiquette for offering y'all this code? Should I fork and submit a PR? Clone and make a feature branch? Happy to contribute however you're comfortable with.
navedmerchant commentedon Jul 6, 2026
Sure please fork and create a PR @rutgoff
Nicramus commentedon Jul 22, 2026
@rutgoff @navedmerchant @CipherChaos there is still an "old" cli in the code but probably it should be removed or something?
rutgoff commentedon Jul 22, 2026
@Nicramus I have a fork of this that I got working, need to find time to write up a proper PR. Hoping to get that done this week.
navedmerchant commentedon Jul 24, 2026
@rutgoff looking forward to the PR!
rutgoff commentedon Jul 25, 2026
Sorry for the delay! Finally found some time to get this knocked out. - Rowena
Bedrock provider support for the CLI is now up as #1015.
It adds
bedrockas a supported CLI provider with the full AWS auth stack (bearer token /AWS_BEDROCK_API_KEY,AWS_PROFILE, direct access keys, and the AWS SDK default credential chain), plus dynamic--providerhelp text and cross-region inference auto-detection forus./eu./apac.model IDs.(Content added by Kenzie, LLM-based artificial assistant, on behalf of Rowena Day.)