ReasonLayer Agent

An AI assistant that runs Claude agents in isolated Docker containers and connects to your team via Slack. Deploy on any Linux server in minutes.

---

How it works

When someone messages the bot in Slack, the message goes into a queue. A polling loop picks it up and spawns a Claude agent inside a Docker container. The agent has access to tools, can call APIs, write files, and do real work — then sends the response back to Slack. Each Slack channel gets its own isolated container with its own memory and filesystem so groups can't see each other's context.

Slack message → SQLite queue → Polling loop → Claude agent container → Response → Slack

The app itself also runs in Docker (via docker compose), so the only thing you need on the host is Docker.

---

Prerequisites

• Docker and Docker Compose — verify with docker compose version
• An Anthropic API key
• A Slack workspace where you can install apps

---

Setup

1. Clone the repo

git clone <your-repo-url>
cd reasonlayer-agent

2. Create your .env file

cp .env.example .env

Open .env and fill in:

ANTHROPIC_API_KEY=your-key-here
SLACK_BOT_TOKEN=xoxb-...
SLACK_APP_TOKEN=xapp-...
ASSISTANT_NAME=Andy

Getting the Slack tokens is covered in the next section.

3. Build the agent image

./container/build.sh

This builds the Docker image that Claude agents run inside. Only needed once, and again after you pull updates.

4. Start

docker compose up -d

Check it's running:

docker compose logs -f

---

Slack Setup

You need to create a Slack app. The slack-app-manifest.json file in this repo pre-configures everything — scopes, events, socket mode — so you don't have to click through the settings manually.

Create the app

1. Go to api.slack.com/apps
2. Click Create New App → From a manifest
3. Select your workspace → Next
4. Paste the entire contents of slack-app-manifest.json → Next → Create

Get your App-Level Token (xapp-...)

1. In your new app, go to Basic Information → scroll down to App-Level Tokens
2. Click Generate Token and Scopes
3. Give it a name (anything), add the connections:write scope, click Generate
4. Copy the token — it starts with xapp-

Get your Bot Token (xoxb-...)

1. In the sidebar, click Install App → Install to Workspace → Allow
2. Copy the Bot User OAuth Token — it starts with xoxb-

Add both tokens to .env, then:

docker compose restart

Add the bot to a channel

In Slack, open the channel you want the bot to join → click the channel name at the top → Integrations → Add apps → search for your bot name and add it.

Register the channel

Get the channel ID from the Slack URL:

https://app.slack.com/client/T.../C0123456789
                                  ^^^^^^^^^^^ this part

Then send this in your main/admin channel:

@Andy join #your-channel-name slack:C0123456789

---

Configuration

All config lives in .env. Restart after changes: docker compose restart

Variable	Required	Default	Description
ANTHROPIC_API_KEY	Yes	—	Anthropic API key
SLACK_BOT_TOKEN	Yes	—	Slack bot token (xoxb-...)
SLACK_APP_TOKEN	Yes	—	Slack app-level token (xapp-...)
ASSISTANT_NAME	No	Andy	The trigger word — messages must start with @Andy
CONTAINER_TIMEOUT	No	1800000	Max agent run time in milliseconds
MAX_CONCURRENT_CONTAINERS	No	5	Max number of agents running in parallel
TZ	No	UTC	Timezone for scheduled tasks

---

Managing the service

docker compose up -d          # Start
docker compose down           # Stop
docker compose restart        # Restart (after .env changes)
docker compose logs -f        # Tail logs

After pulling updates:

docker compose down
./container/build.sh
docker compose up -d

---

Architecture

Single process. One Node.js app handles everything: receiving messages from Slack, queuing them in SQLite, dispatching agents, and sending responses back.

Isolated agents. Each Claude agent runs in its own Docker container. It can only see directories explicitly mounted into it — its group folder, its memory, its IPC channel. It cannot touch other groups, the host filesystem, or your credentials.

Per-group memory. Each Slack channel (group) has its own CLAUDE.md file and working directory that persist between conversations. The agent remembers context, preferences, and ongoing tasks.

IPC via filesystem. The host process and agent containers communicate through mounted directories. The agent writes results to a file, the host picks them up. No network, no ports, no service discovery.

Docker-outside-of-Docker. The ReasonLayer Agent app itself runs in Docker. It spawns agent containers by connecting to the host Docker socket (/var/run/docker.sock). The host just needs Docker — no Node.js, no build tools.

Scheduled tasks. Agents can create recurring tasks (cron-style) that run automatically and message you back with results.