dify

Dify

Dify is an open-source platform for building AI applications, workflows, agents, and RAG pipelines.

What this template deploys

This Akash SDL deploys a minimal self-hosted Dify stack:

• Dify API: langgenius/dify-api:1.14.1
• Dify worker: langgenius/dify-api:1.14.1
• Dify web UI: langgenius/dify-web:1.14.1
• PostgreSQL 15
• Redis 6
• Weaviate vector database
• Dify sandbox
• Dify plugin daemon
• Caddy reverse proxy

The public endpoint is exposed through the caddy service on port 80.

Before deployment

Change all placeholder values in deploy.yaml before using this template in production:

CHANGE_ME_GENERATE_WITH_OPENSSL_RAND_BASE64_42
CHANGE_ME_ADMIN_PASSWORD
CHANGE_ME_DB_PASSWORD
CHANGE_ME_REDIS_PASSWORD
CHANGE_ME_WEAVIATE_API_KEY
CHANGE_ME_SANDBOX_API_KEY
CHANGE_ME_PLUGIN_DAEMON_KEY
CHANGE_ME_PLUGIN_INNER_API_KEY

You can generate secrets with:

openssl rand -base64 42

Make sure that matching values are changed consistently across all services. For example, DB_PASSWORD in api, worker, plugin-daemon, and POSTGRES_PASSWORD in db must be identical.

Deployment

Deploy deploy.yaml through Akash Console or Akash CLI.

After the lease starts, open the URI of the caddy service.

On first launch, Dify should run database migrations through the API service and allow initial setup using INIT_PASSWORD.

Notes and limitations

This is an MVP template for awesome-akash, not a hardened production reference architecture.

Important limitations:

• The template uses local filesystem storage for Dify application files.
• For production usage, S3-compatible storage such as MinIO is recommended.
• The official Dify Docker Compose setup uses an SSRF proxy. This Akash MVP disables the SSRF proxy to avoid mounting external nginx/squid config files.
• The template uses Caddy instead of the official nginx container so routing can be defined inline inside SDL.
• If you add external OAuth providers, email links, or split domains, configure the public URL variables explicitly.

Resource profile

Recommended minimum resources for this template:

• CPU: 10+ total units across services
• RAM: 18+ GiB total across services
• Persistent storage: 80+ GiB total

For small testing, you can reduce storage sizes, but Dify, Weaviate, and PostgreSQL should remain persistent.

Useful links

• Dify documentation: https://docs.dify.ai/
• Dify Docker Compose deployment: https://docs.dify.ai/en/self-host/quick-start/docker-compose
• Dify GitHub repository: https://github.com/langgenius/dify