The pydantic-deep CLI wraps the full [pydantic-deep](https://github.com/vstorm-co/pydantic-deepagents) agent framework into a terminal tool that works like Claude Code or LangChain's Deep Agents CLI. It gives an LLM full access to your local filesystem, shell, planning tools, and skills — so it can autonomously execute complex coding tasks.
Unlike simple chat wrappers, pydantic-deep implements the *deep agent architecture*: planning, subagent delegation, persistent memory, and context management — the same patterns powering Claude Code, Manus AI, and Devin.
Features: 17 slash commands (`/help`, `/compact`, `/context`, `/model`, ...), colored diff viewer for file approvals, visual progress bar, tool call timing, @file mentions, and Ctrl+V image paste.
### Non-Interactive (Benchmark Mode)
```bash
# stdout = response only (clean for piping), stderr = diagnostics
pydantic-deep run "Fix the failing tests in src/"
pydantic-deep run "Create a REST API with FastAPI" --model openai:gpt-4.1
pydantic-deep run "Refactor the auth module" --quiet
```
### Docker Sandbox
Run in an isolated Docker container:
```bash
pydantic-deep run "Build a web scraper" --sandbox --runtime python-web
pydantic-deep chat --sandbox --runtime python-datascience
```
```bash pip install pydantic-deep[cli] pydantic-deep chat ```
## What is this?
The pydantic-deep CLI wraps the full [pydantic-deep](https://github.com/vstorm-co/pydantic-deepagents) agent framework into a terminal tool that works like Claude Code or LangChain's Deep Agents CLI. It gives an LLM full access to your local filesystem, shell, planning tools, and skills — so it can autonomously execute complex coding tasks.
Unlike simple chat wrappers, pydantic-deep implements the *deep agent architecture*: planning, subagent delegation, persistent memory, and context management — the same patterns powering Claude Code, Manus AI, and Devin.
## Usage
### Interactive Chat
```bash pydantic-deep chat pydantic-deep chat --model anthropic:claude-sonnet-4-20250514 ```
Features: 17 slash commands (`/help`, `/compact`, `/context`, `/model`, ...), colored diff viewer for file approvals, visual progress bar, tool call timing, @file mentions, and Ctrl+V image paste.
### Non-Interactive (Benchmark Mode)
```bash # stdout = response only (clean for piping), stderr = diagnostics pydantic-deep run "Fix the failing tests in src/" pydantic-deep run "Create a REST API with FastAPI" --model openai:gpt-4.1 pydantic-deep run "Refactor the auth module" --quiet ```
### Docker Sandbox
Run in an isolated Docker container:
```bash pydantic-deep run "Build a web scraper" --sandbox --runtime python-web pydantic-deep chat --sandbox --runtime python-datascience ```