Install Agent-Zero Without Docker

[facemookdz-tech]

Hi,
can we get Agent-Zero Without Docker installation 🤷‍♀️

[aniruddhaadak80]

A non-Docker path would help a lot of people, especially anyone trying to test quickly on an existing workstation or integrate with a local environment that already has Python, Node, and browser dependencies installed.nnEven if the maintainers do not want to support every host combination, a best-effort manual install guide would still be valuable. For many users the current blocker is not performance, it is simply wanting to understand what the runtime actually depends on outside the container.

[DoThatKarma]

I can see how people want this.. But not for the reason you mention.. If you want to use it locally, you should run it in a docker, or your system would be exposed to the agent(and to the model providers) Docker is a security measure...

And to be fair, docker is really easy to install..

But a ready image for VPS would be nice though.. But letting the agent loose on your own system is not something i recommend.. Better to make a connection some other way(Agent0 has Agent to agent capabilities. MCP(in and out) and a lot more development options.. You can also connect to it with SSH

[RipGPU]

https://github.com/RipGPU/agent-zero
non doker thare in the repo in the zip

[imedox]

Already succed without Docker thanks to AI!

[Draco-Lunaris]

You aren't alone... I'm working through building am installer script to setup agent zero in an lxc instead of using docker. I prefer LXC deployment or bare metal over docker... We seem to be the minority...

[johnnycoderbot-cloud]

Almost all of this will be done by a CLI agent from the cmd prompt

1. Clone the agent zero file to your pc or
2. pull it to your linux.
3. or copy your docker agent zero folder to PC/Linux
4. Run command prompt as administrator.
5. Install a cli agent, ollama, codex, claude, kimi, goose, opencode. Just ask google how
6. Run the Agent.
7. Point the cli agent to your agent zero folder.
8. Tell it to configure agent zero for windows or linux use outside of docker.
9. fix all hard coded paths (Critical) make sure your agentzero folder and directory is where everyhing points to. container is default
10. fix all tool that call to container rfc etc. By pass this (all tools are broken due to this)
11. fix all imports for windows or linux.
12. find and update all references to a container or agent zero main folder that point to wrong location.
13. set up the environment all packages needed to run agent zero with your version of python.
14. smoke test it, smoke test it, smoke test it, smoke test it system wide

Here's your complete Agent-Zero Without Docker installation guide:

---

Phase 1: Repository Setup

Step 1: Clone Agent-Zero

# Windows (PowerShell as Admin) or Linux
git clone https://github.com/agent0ai/agent-zero.git
cd agent-zero

Step 2: Copy Docker Contents (if you have existing container)

If you have a running Docker container with Agent-Zero and want to preserve data:

# Create destination directory
mkdir -p ./a0/usr

# Copy from running container (if available)
docker cp agent-zero:/a0/usr ./a0/
docker cp agent-zero:/a0/logs ./a0/ 2>$null

---

Phase 2: Install CLI Agent

Install one of these CLI agents to help with the migration:

Here's the Ollama CLI quick setup for Agent-Zero:

1. Install Ollama

# macOS/Linux
curl -fsSL https://ollama.com/install.sh | sh

# Windows (PowerShell Admin)
winget install Ollama.Ollama
# OR
# Download from https://ollama.com/download/windows

2. Pull a Coding Model

# Good for coding tasks (Agent-Zero compatible)
ollama pull qwen2.5-coder:14b
ollama pull llama3.2:3b          # Fast, lightweight
ollama pull deepseek-coder:6.7b  # Great for code

3. Quick Ollama CLI Reference

ollama                         # ollama on boarding
ollama list                    # Show downloaded models
ollama run qwen2.5-coder:14b   # Chat interactively
ollama serve                   # Start API server (daemon)
ollama ps                      # Show running models
ollama rm llama3.2:3b          # Delete model
ollama pull qwen2.5-coder:latest  # Update model

or Install OpenAI Codex (Recommended):

npm install -g @openai/codex
# Or visit: https://github.com/openai/codex

or Install Kimi Code (K2.5):

# Visit https://www.moonshot.cn/ (requires API access)
# Or use OpenRouter with Kimi K2.5 model

or Install Goose:

curl -fsSL https://github.com/block/goose/releases/download/stable/download_cli.sh | bash

or Install Claude Code:

npm install -g @anthropic-ai/claude-code

---

Phase 3: Environment Setup - Have CLI Agent Do the Rest

Create Virtual Environment

# Windows
python -m venv .venv
.venv\Scripts\activate

# Linux/macOS
python3 -m venv .venv
source .venv/bin/activate

Install Dependencies

pip install --upgrade pip setuptools
pip install -r requirements.txt
playwright install chromium

Environment Variables

Create .env file in root:

# Required API Keys
API_KEY_OPENAI=your_key_here
API_KEY_ANTHROPIC=your_key_here
API_KEY_OPENROUTER=your_key_here

# A0 Configuration (bypass Docker)
A0_SET_rfc_enabled=false
A0_SET_execution_local=true
A0_SET_docker_enabled=false

# Paths (adjust for your OS)
A0_BASE_PATH=/path/to/agent-zero/a0  # Linux
# A0_BASE_PATH=C:\path\to\agent-zero\a0  # Windows

# Web UI Config
WEB_UI_PORT=50001
WEB_UI_HOST=0.0.0.0

---

Phase 4: Critical Code Modifications

Use your CLI agent (Codex/Kimi/Goose) to make these changes. Point it to your agent-zero folder and give it these specific instructions:

4.1 Fix Hardcoded Docker Paths

Target files: python/helpers/*.py, python/tools/*.py

Search patterns:

• /a0/ → Replace with dynamic path resolution
• /app/ → Replace with project root
• References to docker, container, ssh_rfc

Example fix for path resolution:

# Add to python/helpers/paths.py (create if missing)
import os
from pathlib import Path

def get_a0_base():
    """Get base path for A0 data (works inside and outside Docker)"""
    # Check environment variable first
    if os.getenv('A0_BASE_PATH'):
        return Path(os.getenv('A0_BASE_PATH'))
    
    # Default to project directory
    return Path(__file__).parent.parent.parent / 'a0'

def get_usr_path():
    return get_a0_base() / 'usr'

def get_work_path():
    return get_a0_base() / 'work'

4.2 Bypass RFC Container Calls

Target: python/helpers/intervention_state.py (and related RFC files)

The Problem: RFC tries to call localhost:55080 (or similar) to reach the Docker container . We need to short-circuit this.

Fix instruction for CLI agent:

Find all files containing "rfc_call", "RFCClient", "RemoteFunctionCall", or references to port 55080. 
Create a mock RFC handler that executes functions locally instead of over HTTP.

If the code tries to connect to localhost:55080 or any RFC endpoint, 
make it execute the Python function directly in-process.

Example bypass pattern:

# In python/helpers/rfc_client.py (create/modify)
import os

def rfc_call(function_name, *args, **kwargs):
    """Bypass RFC if running native, else call container"""
    if os.getenv('A0_SET_rfc_enabled', 'false').lower() == 'false':
        # Execute locally - import and call the function directly
        module_path, func_name = function_name.rsplit('.', 1)
        module = __import__(module_path, fromlist=[func_name])
        func = getattr(module, func_name)
        return func(*args, **kwargs)
    else:
        # Original HTTP RFC call to container
        return original_rfc_call(function_name, *args, **kwargs)

4.3 Fix Tool Execution (Critical)

Target: python/tools/code_execution_tool.py, python/tools/terminal_tool.py

Changes needed:

1. Remove Docker/SSH execution paths
2. Force local subprocess execution
3. Fix path mappings

Instruction for CLI agent:

Modify code_execution_tool.py and terminal_tool.py to:
1. Remove any SSH connection attempts to localhost or containers
2. Remove Docker exec calls
3. Ensure subprocess.Popen or subprocess.run uses the local shell (bash/sh for Linux, cmd/powershell for Windows)
4. Change working directories from /a0/ to ./a0/ or use the get_a0_base() function

4.4 Fix Imports for Windows/Linux

Common issues:

• Unix-specific modules (pty, termios, fcntl) - wrap in try/except
• Path separators (/ vs \) - use os.path.join or pathlib.Path
• Signal handling differences

Create compatibility layer (python/helpers/compat.py):

import sys
import os

IS_WINDOWS = sys.platform == 'win32'
IS_LINUX = sys.platform == 'linux'
IS_MAC = sys.platform == 'darwin'

def get_shell():
    if IS_WINDOWS:
        return os.getenv('COMSPEC', 'cmd.exe')
    return os.getenv('SHELL', '/bin/bash')

def get_terminal_size():
    try:
        import shutil
        return shutil.get_terminal_size()
    except:
        return os.terminal_size((80, 24))

4.5 Update All Container References

Search entire codebase for:

• localhost:55080 (RFC port)
• host.docker.internal
• /a0/ (absolute paths)
• container, docker, ssh, rfc
• parse_intervention_command (add this function if missing per your previous issue)

Regex search patterns:

/docker.*exec/
/localhost:\d+/
/\/a0\//
/ssh.*localhost/

---

Phase 5: Specific File Modifications

Based on your previous memory about the missing parse_intervention_command , ensure these specific fixes:

5.1 Fix intervention_state.py

Add the missing function:

def parse_intervention_command(command: str) -> dict:
    """Parse intervention commands from agent output"""
    # Basic implementation - expand as needed
    if not command:
        return {}
    
    parts = command.strip().split()
    return {
        'action': parts[0] if parts else None,
        'args': parts[1:] if len(parts) > 1 else [],
        'raw': command
    }

5.2 Fix run_ui.py and run_cli.py

Ensure they don't assume Docker:

# At start of both files
import os
os.environ.setdefault('A0_SET_rfc_enabled', 'false')
os.environ.setdefault('A0_SET_execution_local', 'true')

---

Phase 6: Directory Structure Setup

Create necessary directories locally:

# Linux/macOS
mkdir -p a0/usr/{chats,files,knowledge,memory,skills,agents}
mkdir -p a0/work
mkdir -p a0/logs
mkdir -p a0/prompts

# Windows PowerShell
New-Item -ItemType Directory -Force -Path a0\usr\chats,a0\usr\files,a0\usr\knowledge,a0\usr\memory,a0\usr\skills,a0\usr\agents
New-Item -ItemType Directory -Force -Path a0\work,a0\logs,a0\prompts

Copy default prompts:

# Copy from repo to a0/prompts
cp -r prompts/* a0/prompts/ 2>/dev/null || echo "Using repo prompts"

---

Phase 7: Testing (Smoke Test Protocol)

Test 1: Basic Import

python -c "from python.helpers import paths; print('Imports OK')"

Test 2: Web UI Launch

python run_ui.py
# Should start on http://localhost:50001
# Check for no RFC connection errors in console

Test 3: CLI Mode

python run_cli.py
# Should start interactive mode without Docker errors

Test 4: Code Execution (Critical)

In the UI/CLI, test:

Execute: print("Hello from native Agent-Zero")

Expected: Output appears without "RFC connection refused" or "Container not found" errors.

Test 5: File System Access

Create a file at ./a0/work/test.txt with content "Native mode works"

Verify: File appears in a0/work/test.txt on host filesystem.

Test 6: Memory/Database

Start a chat, send a message, restart the app.
Verify: Chat history persists (memory is working locally).

---

Phase 8: Platform-Specific Tweaks

Windows-Specific

1. Terminal tool: Ensure it uses cmd.exe or powershell.exe, not /bin/bash
2. Path lengths: Enable long path support in Windows (Group Policy or registry)
3. Antivirus: Exclude the a0/ directory from real-time scanning (improves performance)

Linux-Specific

1. Permissions: Ensure your user owns the a0/ directory
2. Firewall: If accessing remotely, open port 50001
3. Systemd: Optional - create service file for auto-start

---

Troubleshooting Common Issues

Issue	Solution
RFC connection refused	Ensure A0_SET_rfc_enabled=false is set before imports
ModuleNotFoundError: docker	Remove docker imports or install pip install docker
Permission denied on files	Check directory ownership, especially a0/usr
localhost:55080 errors	Search codebase for this port and comment out HTTP calls
Browser automation fails	Ensure Playwright browsers are installed: playwright install chromium
Memory not persisting	Check that ./a0/usr/memory is writable and SQLite can access it

---

Final Checklist

Before declaring success, verify:

• No references to localhost:55080 remain active in code
• No Docker-specific imports crash the app
• Paths resolve to local filesystem, not container paths
• Code execution creates files on host (not in missing container)
• Chat history persists after restart
• Can create, read, update, delete files in a0/work/
• No SSH connection attempts in logs
• Web UI accessible at http://localhost:50001
• CLI mode runs without errors

Security Warning: Running Agent-Zero without Docker removes the security isolation. The agent can now directly access your host filesystem and execute commands. Only run trusted tasks and consider running in a VM for untrusted operations .

Would you like me to elaborate on any specific phase or provide the exact code patches for the RFC bypass layer?

[croge2k]

mind if i ask how you prompted this? i am trying to run all my stuff local so i installed ollama on my laptop and it's going through my instructions saying that it cannot perform the tasks only provide output for ME to do it - I would love to see an AI (or LLM) actually perform something as I have not yet seen any of them do anything useful yet for me - tried agent0 on another machine pointing to ollama:cloud and it did open terminals and such, but the end results were never completed

[johnnycoderbot-cloud]

Get Nvidia NIM they have some good free coding api. Agent zero doesn't like to work on itself (definitely a crash risk). you have to constant make sure all imports are still operation. Domino effects. I used kimi cli/ codex cli in the cmd to fix my agent. Its a nightmare but it is worth it to free your agent from docker. I think if you install some MCPs and skills it can do whatever you ask. If you are in a docker environment ollama needs the correct Base url. just ask google (select AI mode) it was helpful in getting me on the right track.

[imedox]

i make some change on src....only one prompt to fina result output! ❤ using lmstudio & qwen3.5-9b with 162790 context lgth

[RipGPU]

i alredy did it read my coment 100% somting
the zip is inside the repo
https://github.com/RipGPU/agent-zero

[RipGPU]

Hi, can we get Agent-Zero Without Docker installation 🤷‍♀️
IN THE ZIP IN THE REPO
https://github.com/RipGPU/agent-zero

[RipGPU]

ALL ERORS FIXED

[DoThatKarma]

It can work on it self if u clone the repo in project manager! And then work on it as you like :) ©2025 Kunstig Intelligens IT. https://www.kiit.no - Skreddersydde Chatboter og IT Support! Sent from Proton Mail for Android.

…

-------- Original Message --------

On Wednesday, 04/15/26 at 17:34 johnnycoderbot-cloud ***@***.***> wrote: Get Nvidia NIM they have some good free coding api. Agent zero doesn't like to work on itself (definitely a crash risk). you have to constant make sure all imports are still operation. Domino effects. DuckDuckGo removed one tracker. More Report Spam Get Nvidia NIM they have some good free coding api. Agent zero doesn't like to work on itself (definitely a crash risk). you have to constant make sure all imports are still operation. Domino effects. I used kimi cli/ codex cli in the cmd to fix my agent. Its a nightmare but it is worth it to free your agent from docker. I think if you install some MCPs and skills it can do whatever you ask. If you are in a docker environment ollama needs the correct Base url. just ask google (select AI mode) it was helpful in getting me on the right track. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>