Pair Programming — CellCog on Your Machine

Co-work turns any machine into CellCog's workspace. CellCog Desktop acts as a bridge: CellCog's cloud agents coordinate with the desktop app to run commands, read files, and write code directly on the user's machine.

All commands are auto-approved for SDK/agent users — fully autonomous, no manual approval needed.

Prerequisites

This skill requires the cellcog skill for SDK setup and API calls.

# Claude Code, Cursor, Codex + 70 more agents
npx skills add cellcog/skills --skill cellcog

# OpenClaw
openclaw skills install @cellcog/cellcog

How to Use

For your first CellCog task in a session, read the cellcog skill for the full SDK reference — file handling, chat modes, timeouts, and more.

OpenClaw (fire-and-forget):

result = client.create_chat(
    prompt="[your task prompt]",
    notify_session_key="agent:main:main",
    task_label="my-task",
    chat_mode="agent",
    chat_tier="max",
    enable_cowork=True,
    cowork_working_directory="/path/to/project",
)

All agents except OpenClaw (blocks until done):

from cellcog import CellCogClient
client = CellCogClient(agent_provider="openclaw|cursor|claude-code|codex|...")
result = client.create_chat(
    prompt="[your task prompt]",
    task_label="my-task",
    chat_mode="agent",
    chat_tier="max",
    enable_cowork=True,
    cowork_working_directory="/path/to/project",
)
print(result["message"])

Why Co-work?

Your Machine as a Data Source

Your data lives on the user's machine — project files, databases, logs, configs. Instead of uploading everything, enable co-work with a working directory and CellCog agents explore, read, and reason about the data directly. No file size limits, no upload hassle.

CellCog as Your Coding Powerhouse

CellCog agents are among the most capable coding agents available — deep reasoning paired with real execution. Enable co-work and delegate complex coding tasks: build websites, APIs, fix bugs, refactor codebases, set up infrastructure.

CellCog itself is built using this exact co-work capability.

Think of it as a Claude Code or Cursor alternative, backed by CellCog's multi-agent depth and any-to-any engine.

Quick Start

from cellcog import CellCogClient

client = CellCogClient(agent_provider="openclaw")

# 1. Check if desktop app is connected
status = client.get_desktop_status()

# 2. If not connected, get install instructions
if not status["connected"]:
    info = client.get_desktop_download_urls()
    # info contains per-platform URLs + install commands
    # Run the install commands for the user's OS, then:
    # cellcog-desktop --set-api-key <CELLCOG_API_KEY>
    # cellcog-desktop --start

# 3. Create a co-work chat

# OpenClaw agents (fire-and-forget):
result = client.create_chat(
    prompt="Refactor the auth module to use JWT tokens",
    notify_session_key="agent:main:main",  # OpenClaw only
    chat_mode="agent",
    chat_tier="max",
    enable_cowork=True,
    cowork_working_directory="/Users/me/project",
    task_label="refactor-auth",
)

# All other agents (blocks until done):
result = client.create_chat(
    prompt="Refactor the auth module to use JWT tokens",
    chat_mode="agent",
    chat_tier="max",
    enable_cowork=True,
    cowork_working_directory="/Users/me/project",
    task_label="refactor-auth",
)

Desktop App Setup

Call client.get_desktop_download_urls() — returns download URLs and platform-specific install commands for macOS, Windows, and Linux.

After installation:

cellcog-desktop --set-api-key <CELLCOG_API_KEY>
cellcog-desktop --start

The agent can do all of this programmatically — no human interaction needed beyond providing the API key.

Alternatively, ask your human to download CellCog Desktop from cellcog.ai/cowork, open it, and enter their API key.

Desktop CLI Reference

All commands output JSON for easy agent parsing:

Choosing Mode & Tier

Use chat_mode="agent", chat_tier="max" for all coding work — code needs the deepest reasoning tier. The SDK applies "max" automatically whenever enable_cowork=True, so co-work sessions get it even without an explicit tier.

"agent core" is a legacy name that still works forever (the server maps it to Agent max), but new code should pass chat_mode="agent", chat_tier="max".

Agent Team (chat_mode="team") is reserved for deep research — use it only when the task IS research that happens to involve code.

Browse rides Co-work: create_chat(enable_browse=True, browser_profile_id=...) lets the agent drive the user's real Chrome alongside the terminal (profiles discoverable via client.get_browser_status(); enabling Browse auto-enables Co-work server-side).

Error Recovery

If the desktop app disconnects, CellCog auto-fails pending commands with a clear message.

To recover:

cellcog-desktop --stop && cellcog-desktop --start

Then send continue to the chat:

client.send_message(chat_id="abc123", message="continue")

Security

Even with auto-approve, these protections are always active:

• Blocked paths: ~/.ssh, ~/.aws, credential files are inaccessible
• Output redaction: Sensitive data is automatically redacted from command output
• Per-chat scoping: Each chat session is scoped to its working directory

What You Can Build

Co-work enables the full spectrum of development tasks:

• Web development — Build React apps, APIs, landing pages
• Bug fixing — Debug stack traces, fix test failures
• Refactoring — Modernize codebases, improve architecture
• DevOps — Set up CI/CD, Docker configs, infrastructure
• Data pipelines — ETL scripts, database migrations
• Documentation — Generate docs from code, README files

For the best coding experience, also install coding-agent-cellcog:

# Claude Code, Cursor, Codex + 70 more agents
npx skills add cellcog/skills --skill coding-agent-cellcog

# OpenClaw
openclaw skills install @cellcog/coding-agent-cellcog