baish

KO | EN

gitlite — search

#javascript #python #hacktoberfest #ai #llm #golang #machine-learning #deep-learning #docker #nodejs #pytorch #ios #openai #open-source

(★ 14)

A version of bash that knows what you are probably trying to do.

baish 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# baish

`baish` is a fork of GNU Bash (based on Bash 5.3 sources) with an integrated OpenAI-compatible chatbot builtin.

It adds an `ask` builtin (and `?{...}` sugar) that calls an OpenAI-compatible API to answer questions and optionally suggest shell commands.

# Why make this a built-in function rather than an external executable?

- It not only provides the variable expansion feature but can also do intelligent connection caching, making subsequent 'asks' much faster.
- All of the shell variables for controlling the behavior of ask are already built-in and can be set in your .bashrc
- Persistent connection tracking per bash pid would be a heavy lift for any external command.
- As a builtin, `ask` has direct access to bash's internal `OSTYPE` and `HOSTTYPE` variables and sends them to the AI as context, so the model can tailor its command suggestions to your specific OS and CPU architecture without needing to run `uname` or any other discovery commands first.
- I wanted to figure out how to add builtins to bash, and this seemed like an interesting project.
- Because someday, all models will run 100% locally as an API call, so I thought this might be a nice trial run at the problem!
- 
## Build / run / test

### Prerequisites

First, install build dependencies (requires sudo):

```bash
make depends
```

This will install `libcurl4-openssl-dev` which is required for the `ask` and `mcp` builtins. On Debian/Ubuntu/WSL, it also installs and configures locales needed for the test suite. The `make depends` target automatically detects your package manager (apt/brew/pkg) and installs the necessary dependencies.

For manual installation:
- **Debian/Ubuntu/WSL**: `sudo apt-get install libcurl4-openssl-dev locales`
- **macOS**: `brew install curl`
- **FreeBSD**: `sudo pkg install curl`

### Build

Preferred (top-level Makefile):

```bash
make build
make run RUN_ARGS='-c "echo hello"'
make test
```

Manual (bash build system):

```bash
cd bash-source
./configure
make -j4
make tests
```

The resulting shell binary is `bash-source/baish`.

## Install

The top-level Makefile installs the built binary into `$(PREFIX)/bin` (default: `~/.local/bin`):

```bash
make install
# or
make install PREFIX=/usr/local
```

## AI configuration (environment variables)

Required:

- `BAISH_OPENAI_BASE_URL` – least-info-first host/base URL. Any of these work:
  - `llm-host`
  - `llm-host:8000`
  - `http://llm-host/v1`
  - `http://llm-host:8000/v1`
- `BAISH_MODEL` – model name (your server’s model id)

Optional:

- `BAISH_OPENAI_PORT` – port override **only** when `BAISH_OPENAI_BASE_URL` does **not** include an explicit `:port`
- `OPENAI_API_KEY` – bearer token if your server requires auth
- `BAISH_AUTOEXEC` – if non-zero, execute returned commands without prompting (default: `0`) **⚠️ SECURITY WARNING: Setting this to 1 allows automatic execution of LLM-generated commands without confirmation. Only use in trusted environments.**
- `BAISH_FAIL_FAST` – if non-zero, preflight check the LLM `/models` endpoint before sending prompts (helps catch configuration errors early)
- `BAISH_HTTP_TIMEOUT_SECS` – HTTP request timeout in seconds (default: `15`, max: `600`)
- `BAISH_ASK_DEBUG` – if non-zero, enable verbose debug output for ask builtin
- `BAISH_VERBOSE` – if non-zero, enable verbose output (similar to BAISH_ASK_DEBUG)
- `BAISH_ASK_NOTICE` – if non-zero, show informational notices during ask operations

If no port is provided, `baish` will try common HTTP ports (currently `80`, then `8000`).

### Compatibility

`baish` is designed to work with OpenAI-compatible API servers including:
- OpenAI API (chat completions endpoint)
- Local LLM servers (llama.cpp, Ollama, vLLM, etc.) with OpenAI-compatible APIs
- Other providers implementing the OpenAI chat completion format

**Note:** Different API implementations may have varying response formats. The `ask` builtin attempts to parse multiple response formats but works best with servers that strictly follow OpenAI's JSON schema.

### Security Considerations

**⚠️ IMPORTANT SECURITY WARNINGS:**

1. **Command Execution Risk**: The `ask -c` flag and `BAISH_AUTOEXEC=1` setting allow automatic execution of LLM-generated shell commands. This is inherently dangerous:
   - Compromised or misconfigured LLMs could return malicious commands
   - LLM hallucinations might produce destructive commands
   - No sandboxing or validation is performed on returned commands

2. **Recommendations**:
   - **Never** use `BAISH_AUTOEXEC=1` in production or with untrusted LLMs
   - Always review commands before execution (default behavior)
   - Use only with trusted, well-configured LLM endpoints
   - Consider running in a container or restricted environment when testing
   - Be especially careful with commands involving `rm`, `dd`, or system modifications

3. **No Command Timeout**: Executed commands have no built-in timeout. LLM-generated infinite loops or long-running commands will block the shell until interrupted (Ctrl+C).

## Interactive Setup

When you first run `ask` without configuration in an interactive shell, `baish` will offer to guide you through setup:

```bash
$ baish
$ ask "test"
baish: ask: LLM server not configured (missing BAISH_OPENAI_BASE_URL / OPENAI_BASE_URL)
baish: configure the LLM with environment variables:
  export BAISH_OPENAI_BASE_URL=HOST[:PORT][/v1][,HOST...]   # comma/whitespace-separated
  export BAISH_MODEL=MODEL_ID[,MODEL_ID...]                  # comma/whitespace-separated
  export OPENAI_API_KEY=...                     # optional
  export BAISH_FAIL_FAST=1                      # optional preflight
  export BAISH_HTTP_TIMEOUT_SECS=15             # optional transport timeout

Would you like to set this up now? [y/N]: y

=== baish Interactive Setup ===

Enter LLM server URL [localhost:11434]:
Enter API key (press Enter for none):

Probing localhost:11434 for available models...
Found 2 model(s):
  qwen2.5-coder:7b
  llama3.2:3b

Enter model name [qwen2.5-coder:7b]:

Save configuration to ~/.bashrc? [y]: y
Configuration saved to /home/user/.bashrc

=== Setup Complete ===
Variables set in current session:
  BAISH_OPENAI_BASE_URL=localhost:11434
  BAISH_MODEL=qwen2.5-coder:7b

You can now use 'ask' to query the LLM:
  ask "what is 2+2?"
  ask -c "show me the 10 largest files"  # -c to execute commands
```

The interactive setup:
- Offers sensible defaults (Ollama on localhost:11434)
- Probes the server to detect available models
- Optionally persists configuration to `~/.bashrc`
- Sets variables in the current session immediately

## Using the LLM

### `ask` builtin

```bash
ask "how do I list files by size?"
ask -c "show me the 10 largest files"
ask -j "emit JSON only for this question"
```

The model is instructed to return JSON of the form:

```json
{"answer":"...","commands":["..."]}
```

If `commands` are returned:
- default: `baish` prints the commands as a list and does not execute them
- `ask -c`: execute returned commands without prompting (same as `BAISH_AUTOEXEC=1` for the invocation)
- `ask -j`: print the raw JSON response only and do not execute anything

Useful result variables:

- `BAISH_LAST_ANSWER`
- `BAISH_LAST_COMMANDS` (newline-separated)

### `mcp` builtin

The `mcp` builtin enables connecting to external MCP (Managed Control Points) servers for AI-enabled functionality:

```bash
mcp connect localhost:8080     # Connect to an MCP server
mcp list                       # List available commands from the MCP
mcp disconnect                 # Disconnect from the MCP server
```

**Requirements:** `libcurl4-openssl-dev` (or equivalent) must be installed at build time. Run `make depends` to install automatically, or see the Prerequisites section above.

The MCP implementation provides:
- HTTP/HTTPS communication with MCP servers
- Connection state management
- Timeout and error handling (5s connect, 10s operation timeout)

For detailed documentation, see `MCP-IMPLEMENTATION.md`.

### `?{ ... }` syntax sugar

`?{question}` is equivalent to `ask "question"`.

```bash
?{what is the command to find large files in this directory?}
```

## Working examples

```bash
export BAISH_OPENAI_BASE_URL=llm-host
export BAISH_MODEL=MODEL_ID

./bash-source/baish -c '?{what is 2+2?}'
./bash-source/baish -c 'ask "give me a one-liner to show disk usage by directory"'
```

Interactive example:

```bash
export BAISH_OPENAI_BASE_URL=llm-host
export BAISH_MODEL=MODEL_ID

./bash-source/baish
ask "write a safe command to list the 20 largest files under the current directory"
```

If your server is running on a non-default port:

```bash
export BAISH_OPENAI_BASE_URL=llm-host
export BAISH_OPENAI_PORT=8000
export BAISH_MODEL=MODEL_ID

./bash-source/baish -c '?{write a safe rm command to delete ./tmp only}'
```

## Development

### Code Conventions

The baish codebase follows consistent naming conventions:

**Environment Variables:**
- All baish-specific environment variables use `BAISH_*` prefix in uppercase
- Examples: `BAISH_OPENAI_BASE_URL`, `BAISH_MODEL`, `BAISH_AUTOEXEC`
- This follows standard shell conventions for environment variables

**C Functions:**
- All baish-specific functions use `baish_*` prefix in lowercase with underscores
- Examples: `baish_parse_http_base_url()`, `baish_json_escape()`, `baish_slurp_stream()`
- This follows standard C naming conventions and prevents namespace collisions with bash internals

**Result Variables:**
- Shell variables set by builtins use `BAISH_LAST_*` prefix in uppercase
- Examples: `BAISH_LAST_ANSWER`, `BAISH_LAST_COMMANDS`
- Makes it clear these are result variables from the last operation

**Rationale:**
This naming scheme ensures:
- Clear distinction between user-facing shell variables and internal C functions
- No naming conflicts with bash's existing functions and variables
- Consistent, predictable naming for contributors
- Standard C and shell scripting conventions are followed

### Testing

Run the full test suite:
```bash
make test
```

Integration tests with mock servers:
```bash
./tests/integration-test.sh
```

For development and contribution guidelines, see `RELEASING.md` for the release process.

baish

baish 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

baish

baish 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README