An AI assistant with rules you set
HostDock's AI assistant can look into a problem across your servers: it reads logs, checks disks and services, and proposes fixes. You decide which commands it may run on its own. By default that's a list of read-only commands, and anything else waits for you to approve, edit or reject it.
What you can do
Works across servers
Ask about any server by name. The assistant can list your servers, connect, run commands, read and search files, and look at monitoring history.
Approval you control
Choose one of three modes. The approval card names the target server and lets you edit the command before it runs.
Keeps secrets out of prompts
The model sees server names, never addresses, usernames or credentials. Command output is scanned for private keys, tokens and passwords before it is sent.
Command log
Every command the assistant runs is recorded, including the ones that ran automatically. Filter the log by server.
Built in, or your own model
Sign in and use the AI service HostDock provides: no API key, a monthly token allowance, and a choice of models. Or use any OpenAI-compatible API with tool calling; presets cover OpenAI, DeepSeek, Qwen, Kimi, Zhipu, OpenRouter and local Ollama.
Right where you work
Open the panel with ⌘I next to your terminal. Add terminal output to the chat from the right-click menu, or send a container’s logs with one click.
Three execution modes
| Mode | Runs on its own | Asks you first |
|---|---|---|
| Ask every time | Nothing | Every command |
| Safe allowlistDefault | Built-in read-only commands (ls, cat, df, ps, dig, journalctl, systemctl status, docker logs…) plus prefixes you add | Everything else, including any command with redirects, variables or backgrounding |
| Run everything | All commands, after you confirm you understand the risk | Commands that look like they read secrets |
At a glance
| Providers | The AI service HostDock provides (sign in), or any OpenAI-compatible /chat/completions endpoint with tool calling (streaming) |
|---|---|
| Presets | OpenAI, DeepSeek, Qwen, Kimi, Zhipu, OpenRouter, Ollama, or a custom URL |
| Tools | List servers, connect, run command, read file, list folder, system overview, monitoring history, tail file, search files, write file |
| Execution modes | Ask every time · Safe allowlist (default) · Run everything |
| In the chat box | Switch the model for the current chat and change the execution mode without opening settings; send a command from an answer to the terminal without running it |
| Always asks | Commands that look like they read secrets (.ssh/, .env, password files), in every mode |
| Command limits | Non-interactive, 60-second timeout, output trimmed before it reaches the model |
| Redaction | PEM private keys, URL credentials, Authorization headers, password/token fields, sk-, AKIA, gh*_, xox* and JWT tokens |
| Where keys live | API keys are encrypted on your Mac and never synced |
Frequently asked questions
Will the AI assistant run commands without asking me?
Only the ones you allow. In the default “Safe allowlist” mode, about a hundred read-only commands such as ls, df, ps, journalctl and docker ps run automatically, and you can add your own prefixes. Everything else, and anything with redirects, variables or background jobs, waits for your approval. You can also switch to “Ask every time”. Commands that look like they read secrets always ask.
Which AI models can I use with HostDock?
After you sign in, the AI service HostDock provides works without an API key, with a monthly token allowance and a few models to pick from. You can also use any model served through an OpenAI-compatible chat completions API that supports tool calling. HostDock has presets for OpenAI, DeepSeek, Qwen, Kimi, Zhipu, OpenRouter and Ollama, and accepts any other HTTPS endpoint or a local HTTP one.
What does the AI model see about my servers?
The server’s name and group, and the output of commands it runs. It never receives the address, port, username, password or key from your server list, and command output is scanned for common secret formats before it is sent.
More in HostDock
Bring your servers into one window
Free download for macOS 12 Monterey or later.