Rocket + Scale Agentex

This repository is meant for co-development of Agentex agents between Rocket and Scale. It includes a Makefile to spin up the Agentex platform locally as well as the following demo agents:

1. bedrock-temporal-demo — a long-running agent using AWS Bedrock via Temporal workflows
2. example-openai-agent — a synchronous agent using OpenAI models via SGP

Once the platform is running, browse to http://localhost:3000 to interact with the agents.

Quick Start

1. Initialize Submodule

After cloning, initialize the scale-agentex submodule:

git submodule update --init --recursive

2. Set the .env File

Copy .env.example to .env and fill in your credentials:

cp .env.example .env

The following variables are required. Create one set of credentials for non-prod and one for prod:

SGP_API_KEY=
SGP_BASE_URL=
SGP_ACCOUNT_ID=

3. Start the Platform

make platform

This starts the Agentex backend, UI, and all infrastructure (Temporal, Redis, Postgres, MongoDB). Once running:

• Agentex UI: http://localhost:3000
• Agentex API: http://localhost:5003
• Temporal UI: http://localhost:8080

4. Run Agents Locally

Agents run as local Python processes and connect to the platform over the network. Each agent manages its own virtual environment via uv.

Run a single agent:

make agent AGENT=bedrock_temporal_demo
make agent AGENT=example_openai_agent

Run specific agents together:

make agents AGENTS=bedrock_temporal_demo+example_openai_agent

Run all agents:

make agents

The AGENT / AGENTS values are directory names, not agent names defined in manifest.yaml.

Troubleshooting

If you run into issues:

1. Bring down the stack: docker compose down --volumes
2. Prune the Docker system: docker system prune -a
   • Note: This removes all stopped containers and unused images. Use with caution if you have other Docker projects running.

Updating the Submodule

The scale-agentex submodule is pinned to a specific commit.

To update to the latest main:

git submodule update --remote --recursive
git add scale-agentex
git commit -m "Update scale-agentex submodule to latest"

To pin to a specific commit:

cd scale-agentex
git fetch
git checkout <commit-hash>
cd ..
git add scale-agentex
git commit -m "Update scale-agentex submodule to <commit-hash>"

Then rebuild the platform to pick up the changes:

make platform

If the UI is not showing agents after an update, a forced rebuild may be needed due to Docker layer caching:

docker compose --env-file .env -f docker-compose.yaml build --no-cache agentex-ui
docker compose --env-file .env -f docker-compose.yaml up -d agentex-ui

Creating a New Agent

From the repo root, run:

agentex init

This will interactively scaffold a new agent directory with the required files (manifest.yaml, project/acp.py, Dockerfile, etc.). For more information, see the official Agentex documentation

After scaffolding, there are two things to configure manually:

1. Set a Unique Port

Each agent needs a unique port since they all run as local processes sharing the same host. Edit the local_development section in your agent's manifest.yaml:

local_development:
  agent:
    port: 8003  # pick a port not used by another agent

Current port assignments:

Agent	Port
bedrock_temporal_demo	8001
example_openai_agent	8002

Pick the next available port (8003, 8004, etc.) for each new agent.

2. Add SGP Credentials

Add the SGP credential mappings to manifest.yaml under the credentials section:

agent:
  credentials:
    - env_var_name: SGP_API_KEY
      secret_name: sgp-api-key
      secret_key: sgp-api-key
    - env_var_name: SGP_BASE_URL
      secret_name: sgp-base-url
      secret_key: sgp-base-url
    - env_var_name: SGP_ACCOUNT_ID
      secret_name: sgp-account-id
      secret_key: sgp-account-id