Skip to main content

Agent Integration

The Core Idea​

DaoFlow exists because we believe deployment should be as safe for AI agents as version control is for developers. An agent should be able to deploy, diagnose, and rollback — without being able to accidentally break production.

This isn't about "adding AI features" to a hosting platform. It's about building a hosting platform where AI agents are first-class operators with the same safety guarantees humans expect.

Why This Matters​

When an AI coding agent finishes building your feature, the current options are:

  1. Give it SSH access → it can do anything, including rm -rf /
  2. Give it a cloud API → it can spin up $10K in resources by accident
  3. Have a human deploy manually → defeats the purpose of AI agents

DaoFlow is option 4: give the agent scoped, auditable, reversible deployment access.

How It Works​

# 1. Create an agent principal in DaoFlow settings
# 2. Generate a scoped API token
# 3. Configure your AI tool:

daoflow login --url https://deploy.example.com --token dfl_agent_token

# The agent can now safely:
daoflow status --json          # Read infrastructure state
daoflow deploy --dry-run       # Preview without executing
daoflow deploy --yes --json    # Deploy with confirmation
daoflow rollback --dry-run     # Preview rollback
daoflow capabilities --json    # Check what it's allowed to do

Agent Workflows​

WorkflowCommandsRisk Level
Observestatus, logs, doctorNone
Planplan, deploy --dry-run, rollback --dry-runNone
Deploydeploy --yesScoped
Rollbackrollback --yesScoped
Configureenv push --yesScoped

Every command that mutates infrastructure requires both the --yes flag and the correct scope in the token. Without both, the command fails with a structured error telling the agent exactly what scope it needs.

For both service and compose deployments, daoflow deploy ... --dry-run uses the planning lane before execution. Compose previews still inspect the local filesystem for build-context bundle facts first, but the returned plan comes from the control plane and does not mutate anything.

The Safety Contract​

What agents CAN do (with correct scopes):
  ✓ Deploy services
  ✓ Rollback to previous deployments
  ✓ Read logs and diagnose failures
  ✓ Push environment variables
  ✓ View infrastructure state

What agents CANNOT do (ever):
  ✗ Elevate their own permissions
  ✗ Access unmasked secrets
  ✗ Open terminal sessions
  ✗ Override policy guardrails
  ✗ Approve their own requests
  ✗ Delete servers or projects