Skip to main content

Docker Compose Setup

WASP uses Docker Compose to orchestrate 6 services (7 with local LLM). This page explains each service, its configuration, and the volume/network layout.

Services Overview​

agent-redis      — Message broker & KV store
agent-postgres   — Persistent database (18 tables)
agent-core       — Main agent process + dashboard
agent-telegram   — Telegram bridge
agent-nginx      — Reverse proxy + SSL termination
agent-broker     — Docker socket sidecar (privileged)
agent-ollama     — Local LLM server (optional, --profile local-llm)

Service Details​

agent-redis​

image: redis:7-alpine
command: redis-server --appendonly yes --maxmemory 256mb --maxmemory-policy allkeys-lru

Redis serves two purposes:

  1. Event streaming via Redis Streams (events:incoming, events:outgoing)
  2. Key-value storage for state (goals, agents, reminders, API keys, self-model)

AOF persistence is enabled (--appendonly yes). Max memory is 256 MB with LRU eviction. Data is stored in /home/agent/data/redis.

agent-postgres​

image: postgres:16-alpine

PostgreSQL stores all durable data: memories, audit logs, knowledge graph, temporal world model, skill patterns, and all 18 database tables. Data is in /home/agent/data/postgres.

agent-core​

The central service. It:

  • Runs the main event loop (consuming from events:incoming)
  • Hosts the web dashboard on port 8080
  • Runs the scheduler (23 background jobs)
  • Manages the goal engine, agent orchestrator, and skill executor

Key settings:

shm_size: '2gb'          # Required for Chromium browser skill
expose: ["8080"]         # Dashboard only exposed internally (nginx proxies)
SOVEREIGN_MODE: "true"   # Raises MAX_SKILL_ROUNDS to 12

Volumes mounted into agent-core:

Host PathContainer PathPurpose
/home/agent/data/memory/data/memoryMemory snapshots, self-model
/home/agent/data/logs/data/logsStructured logs
/home/agent/config/data/configprime.md operator override
/home/agent/data/screenshots/data/screenshotsBrowser screenshots
/home/agent/data/chat-uploads/data/chat-uploadsUser uploads
/home/agent/data/browser-sessions/data/browser_sessionsChromium profiles
/home/agent/data/skills/data/skillsCustom Python skills
/var/run/docker.sock/var/run/docker.sock:roDocker API (read-only)

agent-telegram​

The Telegram bridge. It:

  • Polls the Telegram Bot API for messages
  • Publishes incoming messages to events:incoming Redis stream
  • Subscribes to events:outgoing for responses
  • Handles media uploads (images, video, documents)
  • Manages live progress editing during multi-round skill execution
TELEGRAM_BOT_TOKEN: ${TELEGRAM_BOT_TOKEN}
TELEGRAM_ALLOWED_USERS: ${TELEGRAM_ALLOWED_USERS:-}  # Comma-separated user IDs

agent-nginx​

Reverse proxy for HTTPS termination. The nginx container:

  • Serves the landing page from /usr/share/nginx/html/
  • Proxies all other traffic to agent-core:8080
  • Handles SSL via Let's Encrypt certificates mounted from the host
ports:
  - "80:80"
  - "443:443"
volumes:
  - /etc/letsencrypt:/etc/letsencrypt:ro

agent-broker​

A privileged sidecar that provides safe Docker access. The agent uses this to:

  • Restart containers (for self-repair)
  • Execute Docker commands via an allowlist

The broker communicates via Redis Streams with an explicit command allowlist — the agent never has direct Docker socket write access.

agent-ollama (optional)​

Local LLM server. Only starts with --profile local-llm:

docker compose --profile local-llm up -d

Memory limit: 12 GB. Models stored in /home/agent/data/ollama.

Networks​

All services share a single bridge network agent-net. Containers communicate using their service names as hostnames (e.g., agent-redis, agent-postgres, agent-core).

Volumes​

All volumes use the local driver with bind mounts to /home/agent/data/:

agent-redis-data    → /home/agent/data/redis
agent-postgres-data → /home/agent/data/postgres
agent-memory        → /home/agent/data/memory
agent-logs          → /home/agent/data/logs
agent-config        → /home/agent/config
agent-backups       → /home/agent/data/backups
agent-shared        → /home/agent/data/shared
agent-screenshots   → /home/agent/data/screenshots
agent-chat-uploads  → /home/agent/data/chat-uploads
agent-browser-sessions → /home/agent/data/browser-sessions
agent-skills        → /home/agent/data/skills
agent-ollama-models → /home/agent/data/ollama

Health Checks​

Both redis and postgres have health checks with condition: service_healthy dependencies. agent-core only starts after both are healthy.

# Monitor health
docker compose ps
docker inspect agent-redis | grep -A5 '"Health"'

Useful Commands​

# Start all services
docker compose up -d

# View logs from all services
docker compose logs -f

# Rebuild and restart agent-core after code changes
docker compose build agent-core && docker compose up -d agent-core

# Scale (postgres/redis only — core must be singleton)
docker compose up -d --scale agent-telegram=0  # Disable Telegram temporarily

# Execute commands inside containers
docker exec -it agent-core bash
docker exec agent-redis redis-cli
docker exec agent-postgres psql -U agent -d agent