Get – tiny agent that gets things from your computer

Developer Water-Run released 'get', an open-source command-line tool that uses a large language model to convert natural language queries into shell commands for retrieving information from a computer. The tool, licensed under AGPL-3.0, includes multiple security layers such as dangerous command blocking and optional manual confirmation to ensure read-only operations.

get is a compact command-line binary tool that calls a Large Language Model LLM via natural language to generate shell commands, then executes them on your device to retrieve any information you need. get is open-sourced under the AGPL-3.0 license on GitHub https://github.com/Water-Run/get . Usage examples: get "IP address of this device" get "code structure in the current directory" get "latest get version at https://github.com/Water-Run/get" Download from GitHub Release https://github.com/Water-Run/get/releases . Just run the installation script directly: python get ready.py Follow the on-screen instructions. If an older installation is found, the installer asks whether to keep the existing configuration while replacing only the binary. Once done, run get version to verify the installation. Before use, you must configure at least the LLM-related parameters. get is compatible with the OpenAI API specification. Use the following commands to configure: get set model your-model-name get set url your-api-endpoint get set key your-api-key After configuration, run get isok to verify. To clear a configuration value, leave the value empty. For example, to clear the key: get set key Usage is straightforward: get "your question" get is designed to perform read-only operations only. Every generated command goes through multiple layers of security validation before execution built-in dangerous command regex blocking, plus optional double-check and manual confirmation . Integer options accept false to disable the feature entirely equivalent to 0 . | Option | Description | Value Type | Default | |---|---|---|---| key | LLM API key | String | Empty | url | LLM API endpoint URL | String URL | https://api.xiaomimimo.com/v1 | model | LLM model name | String | mimo-v2.5-pro | manual-confirm | Require manual confirmation before execution | true / false | false | double-check | Enable secondary safety review of commands | true / false | true | instance | Use single-call fast mode | true / false | false | timeout | Timeout per API request | Positive integer s / false | 300 | max-token | Maximum token consumption per request | Positive integer / false | 20480 | max-rounds | Maximum loop rounds in agent mode | Positive integer / false | 3 | command-pattern | Regex pattern to block commands | Regex string | Built-in blocklist omit value to restore; "" to disable | system-prompt | Custom system prompt | String | Empty | shell | Shell used to execute commands | String | Windows: powershell ; Linux: bash | log | Log each request and execution | true / false | true | hide-process | Hide intermediate steps | true / false | false | cache | Enable response caching | true / false | true | cache-expiry | Cache entry expiry in days | Positive integer days / false | 30 | cache-max-entries | Maximum number of cache entries to retain | Positive integer / false | 1000 | cache-trigger-threshold | Number of prior executions required before cache decision | Positive integer / false | 1 | log-max-entries | Maximum number of log entries to retain | Positive integer / false | 1000 | vivid | Enable vivid output mode semantic colors and progress spinners | true / false | true | Examples of disabling integer options: get set timeout false get set cache-expiry false get set log-max-entries false command-pattern is matched against the command string before shell expansion. Blocking a literal filename can be bypassed by globbing for example, cat .txt , so reliable policies should target command verbs cat , head , cp , etc. instead. The following flags can override persistent configuration for a single query, placed after the query string: | Flag | Description | |---|---| --no-cache | Bypass cache for this query | --cache | Force cache use for this query | --manual-confirm | Require confirmation before execution | --no-manual-confirm | Skip confirmation prompt | --double-check | Enable secondary safety review | --no-double-check | Skip secondary safety review | --instance | Fast single-call mode | --no-instance | Multi-round agent mode | --hide-process | Hide intermediate process output | --no-hide-process | Show intermediate process output | --vivid | Enable vivid output mode | --no-vivid | Use plain text output mode | --model <name | Override the LLM model for this query | --timeout <seconds | Override the request timeout this query | Examples: get "disk usage" --no-cache get "list files" --model mimo-v2.5-pro --vivid The config command is used to inspect the current configuration. get config Show all current configuration get config --reset Reset all configuration to defaults get config --<option Show the current value of a single option --<option accepts all option names from the set table above. get config --key shows whether a key is configured. The stored value is encrypted and cannot be retrieved. Disabled integer options display false . get config --command-pattern View active pattern full regex when default get set command-pattern Restore built-in default get set command-pattern "" Disable pattern filtering get uses a deferred-decision caching mechanism. Queries are not cached on first execution. Cache strategy classification is triggered only after a query has been seen at least cache-trigger-threshold times default 1 , which means classification on the second run , or when --cache is explicitly passed. During this decision step, get shows checking whether to cache... as progress text when process output is not hidden . Five strategies are supported: GLOBAL COMMAND cache the command globally, re-execute on hit , GLOBAL RESULT cache the output globally, return directly , CONTEXT COMMAND cache the command for the current directory context, re-execute on hit , CONTEXT RESULT cache the output for the current directory context, return directly , or NOCACHE do not cache . Global entries work across any working directory; context entries are tied to the directory where the query was originally executed. When hide-process is enabled, get suppresses its own intermediate progress and warning text including cache checks and model advisory warnings , and only prints the final result or required errors. When config cache is false , get emits a warning in normal process mode: warning: cache is disabled in config; all cache logic is bypassed . get cache Show cache status get cache --clean Clear all cache entries and seen records get cache --unset "system version" Remove cache entries matching the query get set cache false Disable caching disables all cache logic get records each query execution to a local file. When the number of entries exceeds the log-max-entries limit, the oldest entries are automatically removed. get log Show log status get log --clean Clear all log entries get set log false Disable logging get get Show basic information about get name, version, author, ... get get --intro Show introduction get get --version Show version equivalent to get version get get --license Show license identifier get get --github Show GitHub link get isok Verify that the current configuration is usable get help Show usage help Config file : Linux ~/.config/get/config.json / Windows %APPDATA%/get/config.json API key : Linux ~/.config/get/key permissions 0600 / Windows %APPDATA%/get/key DPAPI encrypted Execution log : Linux ~/.config/get/get.log / Windows %APPDATA%/get/get.log Response cache : Linux ~/.config/get/cache.json / Windows %APPDATA%/get/cache.json 0 : Completed successfully 1 : Configuration error, LLM communication failure, security check rejection, or general error 130 : Interrupted by Ctrl+C SIGINT N : When a generated command exits with a non-zero code, that exit code is passed through as get 's exit code