Agency

Deploy autonomous AI coding agents on GCP, AWS, Railway, or locally via Docker

One command to launch an AI agent that writes code for you:

agency-quickdeploy launch "Build a Python CLI that converts CSV to JSON"

The agent runs autonomously in a VM or container, writing code until the task is complete.

Choose Your Path

I want to try it locally (free, no cloud)

pip install -e .
export ANTHROPIC_API_KEY=sk-ant-api03-...  # Get from console.anthropic.com
agency-quickdeploy launch "Build a calculator CLI" --provider docker

I want cloud VMs with SSH access

pip install -e .
export ANTHROPIC_API_KEY=sk-ant-api03-...
export QUICKDEPLOY_PROJECT=my-gcp-project   # GCP project ID
gcloud auth application-default login
agency-quickdeploy launch "Build a REST API"

I want to use my Claude subscription (not pay-per-token)

# Generate OAuth token (one-time, requires browser)
claude setup-token
# Returns: sk-ant-oat01-...

export CLAUDE_CODE_OAUTH_TOKEN=sk-ant-oat01-YOUR-TOKEN
agency-quickdeploy launch "Build an app" --auth-type oauth --provider docker

Quick Start by Provider

Docker (Local, Free)

Run agents on your own machine. No cloud costs.

# Prerequisites: Docker installed

# 1. Install
pip install -e .

# 2. Set API key
export ANTHROPIC_API_KEY=sk-ant-api03-...

# 3. Initialize (pulls agent image)
agency-quickdeploy init --provider docker

# 4. Launch
agency-quickdeploy launch "Build a todo app with SQLite" --provider docker

# 5. Monitor
agency-quickdeploy status <agent-id> --provider docker
agency-quickdeploy logs <agent-id> --provider docker
docker exec -it <agent-id> bash    # Shell into container

# 6. Stop
agency-quickdeploy stop <agent-id> --provider docker

GCP (Full VMs with SSH)

Full Linux VMs with SSH access and GCS storage.

# Prerequisites: gcloud CLI installed

# 1. Install
pip install -e .

# 2. Configure
export ANTHROPIC_API_KEY=sk-ant-api03-...
export QUICKDEPLOY_PROJECT=your-gcp-project

# 3. Authenticate
gcloud auth login
gcloud auth application-default login

# 4. Launch
agency-quickdeploy launch "Build a REST API with FastAPI"

# 5. Monitor
agency-quickdeploy status <agent-id>
agency-quickdeploy logs <agent-id>
gcloud compute ssh <agent-id> --zone=us-central1-a    # SSH in

# 6. Stop
agency-quickdeploy stop <agent-id>

# Pro tip: Use spot instances for 60-90% cost savings
agency-quickdeploy launch "Build something" --spot

AWS (EC2 + S3)

EC2 instances with S3 storage.

# Prerequisites: AWS CLI configured

# 1. Install
pip install -e .

# 2. Configure
export ANTHROPIC_API_KEY=sk-ant-api03-...
export AWS_REGION=us-east-1

# 3. Launch
agency-quickdeploy launch "Build a Lambda function" --provider aws

# 4. Monitor
agency-quickdeploy status <agent-id> --provider aws
agency-quickdeploy logs <agent-id> --provider aws

# 5. Stop
agency-quickdeploy stop <agent-id> --provider aws

# Pro tip: Use spot instances
agency-quickdeploy launch "Build something" --provider aws --spot

Railway (Fast Containers)

Lightweight containers with fast startup.

# Prerequisites: Railway account

# 1. Install
pip install -e .

# 2. Configure
export ANTHROPIC_API_KEY=sk-ant-api03-...
export RAILWAY_TOKEN=...    # Get from railway.com/account/tokens

# 3. Launch
agency-quickdeploy launch "Build a Discord bot" --provider railway

# 4. Monitor
agency-quickdeploy status <agent-id> --provider railway
agency-quickdeploy logs <agent-id> --provider railway

# 5. Stop
agency-quickdeploy stop <agent-id> --provider railway

Provider Comparison

	Docker	GCP	AWS	Railway
Cost	Free	~$0.02/hr	~$0.02/hr	Per-usage
Startup	~5s	~2-3 min	~2-3 min	~30s
SSH	`docker exec`	Yes	Via key	No
Spot/Preemptible	N/A	Yes	Yes	No
Best for	Development	Production	AWS users	Quick tests

Using Your Claude Subscription (OAuth)

Instead of paying per-token with an API key, use your Claude Code subscription:

Step 1: Generate Token

# On a machine with a browser
claude setup-token
# Opens browser, you authenticate
# Returns: sk-ant-oat01-...

Step 2: Use Token

Simple (Environment Variable):

export CLAUDE_CODE_OAUTH_TOKEN=sk-ant-oat01-YOUR-TOKEN
agency-quickdeploy launch "Build an app" --auth-type oauth --provider docker

GCP Secret Manager (More Secure):

# Store token
echo '{"claudeAiOauth":{"accessToken":"sk-ant-oat01-YOUR-TOKEN"}}' | \
    gcloud secrets create claude-oauth-credentials --data-file=-

# Launch (reads from Secret Manager automatically)
agency-quickdeploy launch "Build an app" --auth-type oauth

CLI Reference

Commands

Command	Description
`launch "PROMPT"`	Start a new agent
`status <id>`	Check agent status
`logs <id>`	View agent logs
`list`	List all agents
`stop <id>`	Stop and delete agent
`init`	Verify configuration

Common Options

agency-quickdeploy launch "PROMPT" [OPTIONS]

  -p, --provider [gcp|aws|docker|railway]   # Deployment target (default: gcp)
  -a, --auth-type [api_key|oauth]           # Authentication (default: api_key)
  -n, --name TEXT                           # Custom agent name
  -r, --repo URL                            # Git repo to clone
  -b, --branch NAME                         # Git branch
  --spot                                    # Use spot instances (GCP/AWS)
  --shutdown                                # Auto-shutdown on completion (default: stays running)
  -m, --max-iterations N                    # Limit iterations (0=unlimited)

Environment Variables

Required

Variable	Provider	Description
`ANTHROPIC_API_KEY`	All	API key (for pay-per-token)
`CLAUDE_CODE_OAUTH_TOKEN`	All	OAuth token (for subscription)
`QUICKDEPLOY_PROJECT`	GCP	GCP project ID
`RAILWAY_TOKEN`	Railway	Railway API token

Optional

Variable	Default	Description
`QUICKDEPLOY_PROVIDER`	`gcp`	Default provider
`QUICKDEPLOY_ZONE`	`us-central1-a`	GCP zone
`AWS_REGION`	`us-east-1`	AWS region
`AGENCY_DATA_DIR`	`~/.agency`	Docker data directory

How It Works

Launch: Creates a VM/container with your prompt
Initialize: Agent creates feature_list.json breaking down the task
Execute: Agent implements features one-by-one across sessions
Sync: Logs and code sync to cloud storage
Complete: VM/container terminates (or stays up with --no-shutdown)

Based on Anthropic's autonomous-coding pattern.

agentctl (Advanced)

For managing multiple agents with a central server:

# Install with server support
pip install -e ".[server]"

# Start master server
uvicorn agentctl.server.app:app --port 8000

# For development (auto-reload on changes)
uvicorn agentctl.server.app:app --reload --port 8000

# Point CLI to server
export AGENTCTL_MASTER_URL=http://localhost:8000

# Use CLI
agentctl run "Build an API"
agentctl list
agentctl ssh my-agent
agentctl logs my-agent --follow
agentctl tell my-agent "Also add tests"

See CLAUDE.md for full agentctl documentation.

Troubleshooting

Agent stuck initializing

# Check status
agency-quickdeploy status <agent-id>

# View logs
agency-quickdeploy logs <agent-id>

# SSH in (GCP)
gcloud compute ssh <agent-id> --zone=us-central1-a
tail -f /var/log/agent.log

No logs appearing

Logs sync every 60 seconds. For immediate logs:

# GCP
gcloud compute ssh <agent-id> --command="tail -50 /var/log/agent.log"

# Docker
docker logs <agent-id>

OAuth not working

Token must start with sk-ant-oat (not sk-ant-api)
Use --auth-type oauth flag
For GCP, ensure Secret Manager permissions are set

Development

# Install dev dependencies
pip install -e ".[server,dev]"

# Run tests
python -m pytest

# Run with coverage
python -m pytest --cov=agency_quickdeploy --cov=agentctl

Documentation

Document	Purpose
CLAUDE.md	Technical reference (for developers & AI agents)
BACKLOG.md	Known issues and planned work
docs/deployment/	Detailed deployment guides
docs/api/API.md	REST API reference

License

MIT License - see LICENSE

Links

Anthropic Console - Get API key
Claude Code - Generate OAuth token with claude setup-token
Railway - Get Railway token

Name		Name	Last commit message	Last commit date
Latest commit History 110 Commits
.github/workflows		.github/workflows
agency_quickdeploy		agency_quickdeploy
agent-downloads/agent-20260107-000102-e80f73c4		agent-downloads/agent-20260107-000102-e80f73c4
agent-runner		agent-runner
agentctl		agentctl
docs		docs
scripts		scripts
shared		shared
tests		tests
.env.example		.env.example
.gitignore		.gitignore
BACKLOG.md		BACKLOG.md
CLAUDE.md		CLAUDE.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
pyproject.toml		pyproject.toml
railway.toml		railway.toml
requirements-dev.txt		requirements-dev.txt
requirements.txt		requirements.txt

License

wesleyzhao/agency

Folders and files

Latest commit

History

Repository files navigation

Agency

Choose Your Path

I want to try it locally (free, no cloud)

I want cloud VMs with SSH access

I want to use my Claude subscription (not pay-per-token)

Quick Start by Provider

Docker (Local, Free)

GCP (Full VMs with SSH)

AWS (EC2 + S3)

Railway (Fast Containers)

Provider Comparison

Using Your Claude Subscription (OAuth)

Step 1: Generate Token

Step 2: Use Token

CLI Reference

Commands

Common Options

Environment Variables

Required

Optional

How It Works

agentctl (Advanced)

Troubleshooting

Agent stuck initializing

No logs appearing

OAuth not working

Development

Documentation

License

Links

About

Resources

License

Contributing

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors 2

Uh oh!

Languages

Packages