Introducing DeepAgents CLI

Introducing DeepAgents CLI

3 min read

By Vivek Trivedy

We're excited to introduce DeepAgents CLI for coding, research, and building agents with persistent memory. Now you can easily create and run custom DeepAgents directly from the terminal. It supports:

  • Read, write, and edit files in your project
  • Execute shell commands with human approval
  • Search the web for current information
  • Make HTTP requests to APIs
  • Learn and remember information across sessions
  • Plan tasks with visual todo lists

Installation

Install DeepAgents with CLI support:


pip install deepagents-cli

Or if you're using uv:


uv pip install deepagents-cli

Quick Start

1. Set Up Your API Keys

DeepAgents CLI supports both Anthropic (Claude) and OpenAI models. Anthropic Claude Sonnet 4 is the default model and Tavily is used for web search. Add these to your .env file in your project root, and DeepAgents will automatically load them:


export ANTHROPIC_API_KEY=your_api_key_here
export OPENAI_API_KEY=your_api_key_here
export TAVILY_API_KEY=your_tavily_key_here

2. Launch the CLI

Start DeepAgents in your project directory:


deepagents

Or, if you’re using uv:

uv run deepagents

3. Your First Task

Try asking the agent to help with a simple task:


You: Add type hints to all functions in src/utils.py

The agent will:

  1. Read the file
  2. Analyze the functions
  3. Show you a diff of proposed changes
  4. Ask for your approval before writing

There's also an option to Auto-Accept Edits to speed up development

Learning Through Memory

One of DeepAgents' most powerful features is its persistent memory system. The agent can learn information and recall it across sessions. Each agent stores its knowledge in ~/.deepagents/AGENT_NAME/memories/:

By default, if you spin up DeepAgents it will create an agent with the name agent and use that by default. You can change the agent used (and therefor what memories are used) by specifying an agent name, eg deepagents --agent foo. See next section for more details.

The agent automatically follows a Memory-First Protocol:

  1. During Research - Checks /memories/ for relevant knowledge
  2. Before answering - Searches memory files in case of uncertainty
  3. When learning - Saves new information to /memories/

Example: Teaching API Patterns


You: Remember that our API endpoints follow this pattern:
- Use /api/v1/ prefix
- All POST requests return 201 on success
- Error responses include a "code" and "message" field

Save this as our API conventions.

Agent: I'll save these API conventions to memory.
⚙ write_file(/memories/api-conventions.md)

Because this memory is persistent, the agent can use this information across future conversations.

You: Create a new endpoint for user registration
Agent: Based on our API conventions, I'll create an endpoint at
/api/v1/users that returns 201 on success and follows
our error format.
⚙ read_file(/memories/api-conventions.md)
⚙ write_file(src/routes/users.py)

Memory Best Practices

1. Use descriptive filenames ✓ /memories/deployment-checklist.md ✗ /memories/notes.md

2. Organize by topic

/memories/
├── backend/

│ ├── tools_to_use.md

│ └── api-design.md

├── frontend/

│ ├── component-patterns.md

└── security-setup.md

3. Verify saved knowledge Because memory is just a set of files, you can always inspect and validate its content manually or with the agent.

You: Check what you know about our database

Agent: Let me check my memories...
⚙ ls /memories/
⚙ read_file(/memories/backend/database-schema.md)

Based on my memory, we use PostgreSQL with these tables...

You can also inspect the memory files manually by just looking at ~/.deepagents/AGENT_NAME/memories/

Managing Multiple Agents

You can create specialized agents for different projects or roles: From the DeepAgents CLI you can list existing agents, create new agents, or reset an agent to its default state (system prompts, memories, etc).

deepagents --list


deepagents --agent backend-dev


deepagents --reset backend-dev

Get Started Today

Get started with DeepAgents and the DeepAgent CLI today! We're excited to see what you build.

Join the community and contribute: