智能代理管理

Prerequisites

You need Docker. Don't have it? Install Docker Desktop for Mac/Windows or Docker Engine for Linux. Check: run docker --version in a terminal; if you see a version, you're set.

Docker is the only supported way to run AgentManager. Running the server or UI outside Docker is unsupported and unsafe.

Prerequisites

• GCP project with billing enabled
• gcloud CLI authenticated and configured
• terraform CLI installed
• Docker (if building locally)

Deploy to Cloud Run in 5 Steps

Step 1: Build and push the container image

```bash

Option A: Build remotely with Cloud Build (recommended, no local Docker needed)

gcloud builds submit \ --tag $REGION-docker.pkg.dev/$PROJECT_ID/agent-manager/agent-manager:latest \ --project=$PROJECT_ID --region=$REGION

Option B: Build locally with Docker

docker build -t $REGION-docker.pkg.dev/$PROJECT_ID/agent-manager/agent-manager:latest . docker push $REGION-docker.pkg.dev/$PROJECT_ID/agent-manager/agent-manager:latest ```

Step 3: Deploy infrastructure with Terraform

terraform init
terraform plan    # Preview changes
terraform apply   # Deploy

This creates: - Cloud Run service - 32GB RAM, 8 CPU, autoscaling (min 0, max 1 instance) - GCS bucket - Persistent storage for agent state and shared context - Secret Manager secrets - OpenRouter key, API key, JWT secret, MCP credentials - Service account - Minimal IAM permissions (Cloud Run invoker, GCS admin, Secret Manager accessor) - IAM auth - No public access; requires authenticated users - Cloud Monitoring alerts - Error rate, p99 latency, crashes, memory, CPU

Redeploy to pick up new secrets

gcloud run services update agent-manager --region=$REGION ```

Quick Start

1. Clone the repository

   git clone https://github.com/simonstaton/AgentManager.git AgentManager
   cd AgentManager
   

2. Configure environment variables

   cp .env.example .env
   

3. Start the app

   npm run docker:local
   

1. Open the UI — Go to http://localhost:8080 and log in with your API_KEY.

Your data (repos, shared context, logs) is stored in a Docker volume and survives restarts. Full steps and troubleshooting: Run locally with Docker.

Usage Examples

Step 2: Configure Terraform variables

cd terraform
cp terraform.tfvars.example terraform.tfvars

Edit terraform.tfvars with your values: - project_id - Your GCP project ID - region - Deployment region (e.g. us-central1) - api_key - Your UI login password - openrouter_api_key - Get from openrouter.ai/keys - jwt_secret - Any random 32+ character string - Optional: github_token, figma_token, linear_api_key, notion_api_key, etc.

Config

Settings

Why Claude Code CLI, not the Anthropic SDK?

The SDK gives you chat completions with tool calling. Claude Code gives you a complete coding agent with file editing, bash execution, git, MCP, session resumption and sub-agent delegation, all maintained by Anthropic. AgentManager runs these agents rather than trying to rebuild them from scratch.

Claude Code's --output-format stream-json gives typed JSON events (not terminal scraping) that the platform parses for real-time UI streaming, state tracking and cost calculation. New capabilities Anthropic adds to Claude Code show up in AgentManager without any work on my end.

Layer 6: Remote kill when API is unreachable

echo '{"killed":true,"reason":"emergency"}' | gsutil cp - gs://your-bucket/kill-switch.json ```

API Reference

<details> <summary>Click to expand full API reference</summary>

GitHub Integration

Setting GITHUB_TOKEN enables three things for agents: 1. gh CLI - create PRs, manage issues, query repos 2. git push/git fetch - credential helper is configured automatically on startup via gh auth setup-git 3. GitHub MCP server - structured tool access to GitHub's API

Creating a token

Option A: Fine-grained token (recommended)

Go to GitHub Settings > Fine-grained tokens: - Token name: agent-manager - Expiration: 90 days (or custom) - Repository access: "Only select repositories" and pick the repos agents should access - Permissions: - Contents - Read and write - Pull requests - Read and write - Metadata - Read-only (auto-selected)

Option B: Classic PAT

Go to GitHub Settings > Tokens (classic): - Scopes: repo (required), workflow (optional)

Configuring the token

When running with Docker locally — add to .env:

GITHUB_TOKEN=github_pat_xxxxx

Production (Cloud Run) - add to terraform/terraform.tfvars:

github_token = "github_pat_xxxxx"

Quick update without Terraform - update the secret directly:

echo -n "github_pat_new_token_here" | gcloud secrets versions add github-token --data-file=- --project=$PROJECT_ID
gcloud run services update agent-manager --region=$REGION --project=$PROJECT_ID