Resource Reference

All 14 managed resources. Each entry shows: HCL, the most important arguments, what's computed, and the immutable / sensitive fields.

Conventions used throughout:

• Required — must be set in HCL.
• Optional — has a default or is genuinely optional.
• Computed — never set in HCL; populated from the API after apply.
• Sensitive — masked in plan output and terraform output. Stored in state; protect your state.
• Immutable — changing forces destroy + recreate (RequiresReplace).

---

agentcy_source

Ingestion connector. The set of valid source_type values is exposed by the agentcy_agent_types-equivalent connector catalog at GET /api/v1/sources/catalog.

resource "agentcy_source" "github" {
  source_type = "github_pat"      # Required, immutable
  name        = "main-monorepo"   # Required
  config      = jsonencode({      # Required, sensitive
    token        = var.gh_token
    repositories = ["acme/monorepo"]
  })
  schedule    = "0 */6 * * *"     # Optional cron
  visibility  = "organization"    # Optional
  realm       = "engineering"     # Optional, defaults from source_type
}

Argument	Notes
source_type	Required, immutable. One of github_pat, aws, gcp, vercel, csv, sql, mongodb, etc.
config	Required, sensitive. Per-type JSON object — see the connector docs.
schedule	Optional cron expression for auto-sync. Empty string disables.

Computed: id, status, created_at.

---

agentcy_settings

Singleton — one per organization. Create and Update both PUT the desired state; Delete is a no-op.

resource "agentcy_settings" "this" {
  organization_name     = "Acme"
  llm_provider          = "openai"
  llm_model             = "gpt-4o"
  llm_base_url          = "https://api.openai.com/v1"
  embedding_model       = "all-MiniLM-L6-v2"
  auto_embed_on_ingest  = true
  approval_timeout_secs = 300
}

id is always org-settings.

---

agentcy_realm

Graph namespace. The default realm exists automatically and cannot be deleted.

resource "agentcy_realm" "engineering" {
  name         = "engineering"        # Required, immutable, lowercase a-z0-9-_, ≤100
  display_name = "Engineering Team"
  description  = "Eng-only graph data"
  color        = "#3b82f6"
}

---

agentcy_agent

Typed agent (enrichment, anomaly detection, relationship discovery, …).

resource "agentcy_agent" "enricher" {
  agent_type = "enrichment"           # Required, immutable
  name       = "team-enricher"        # Required
  config     = jsonencode({})         # Optional
  enabled    = true                   # Optional, defaults true
}

Use the agentcy_agent_types data source for the catalog of valid agent_type values.

---

agentcy_webhook

Inbound webhook endpoint. The secret_token is plaintext only at create time and is recorded in state as a sensitive computed value.

resource "agentcy_webhook" "alerts" {
  name             = "alerts-hook"
  description      = "PagerDuty alerts"
  payload_template = "{ \"alert\": {{json .body}} }"
  system_prompt    = "Triage incoming alerts and route to oncall."
}

output "alert_url"    { value = agentcy_webhook.alerts.url }
output "alert_secret" { value = agentcy_webhook.alerts.secret_token, sensitive = true }

Computed: id, slug, secret_token, url, created_at.

To rotate the secret: destroy and recreate. There's no in-place rotation today.

---

agentcy_skill

Versioned, named, scope-tagged content blob agents can call.

resource "agentcy_skill" "summarize" {
  name        = "Summarize PR"
  slug        = "summarize-pr"        # Required, immutable
  description = "Summarize a GitHub PR"
  version     = "0.1.0"
  tags        = ["github", "summary"]
  content     = file("${path.module}/skills/summarize-pr.md")
  scope       = "org"                 # Optional, immutable; "org" or "user"
}

---

agentcy_memory

Persistent memory entries.

resource "agentcy_memory" "house_style" {
  content     = "Default response tone: concise, technical, no fluff."
  title       = "House style"
  memory_type = "guideline"
  tags        = ["style", "tone"]
  realm       = "engineering"         # Optional, immutable
}

The importance field (0.0–1.0) is intentionally not exposed in v1 — the backend stores it as 32-bit float, which round-trips with precision drift. Manage importance via the API directly.

---

agentcy_api_key

Programmatic API key. The plaintext key is returned only at create and is persisted in state as a sensitive computed value. Delete is a soft revoke (the row stays in the DB with active = false).

resource "agentcy_api_key" "ci" {
  name            = "ci-pipeline"
  permissions     = ["agent:manage", "source:manage"]
  source_ids      = [agentcy_source.repo.id]   # Optional scoping
  expires_in_days = 90                         # Immutable; rotate by recreating
}

output "ci_key" { value = agentcy_api_key.ci.key, sensitive = true }

---

agentcy_task

Cron-scheduled task that runs an agent.

resource "agentcy_task" "morning_summary" {
  name             = "morning-summary"
  cron_expression  = "0 9 * * 1-5"
  timezone         = "UTC"
  message_template = "Summarize yesterday's PRs."
  system_prompt    = "Be concise."
  response_channel = "slack:#eng"
  source_ids       = [agentcy_source.repo.id]
  is_active        = true
  trigger_type     = "schedule"        # Optional, immutable. "schedule" (default) or "webhook".
}

For webhook tasks (trigger_type = "webhook"), webhook_slug and webhook_url are computed.

---

agentcy_role

RBAC role. Built-in roles (is_builtin = true) cannot be modified.

resource "agentcy_role" "viewer" {
  name        = "viewer"
  description = "Read-only access"
  permissions = ["graph:read", "agent:manage"]
}

Use agentcy_permissions to discover valid permission strings.

---

agentcy_policy

Rego policy. Provide the body via rego_code (inline) or rego_file (path on the runner) — exactly one is required. The provider validates the Rego against the backend's /policies/policies/validate endpoint at plan time, so syntactic errors fail before apply.

resource "agentcy_policy" "business_hours" {
  name         = "business-hours-only"
  package_path = "agentcy.business_hours"
  rego_code    = <<-EOT
    package agentcy.business_hours
    default allow := true
    deny[msg] if {
      input.action == "write"
      hour := time.clock([time.now_ns(), "UTC"])[0]
      hour < 9
      msg := "Writes blocked before 09:00 UTC"
    }
  EOT
  enabled = true
}

# Or load from a file:
resource "agentcy_policy" "from_file" {
  name         = "tenant-isolation"
  package_path = "agentcy.tenant_iso"
  rego_file    = "${path.module}/policies/tenant_iso.rego"
}

---

agentcy_policy_source

Git-backed source for Rego policies.

resource "agentcy_policy_source" "shared" {
  name        = "shared-policies"
  source_type = "git_repo"            # Required, immutable. "git_repo" or "inline".
  git_url     = "git@github.com:acme/shared-policies.git"
  git_branch  = "main"
  git_path    = "policies/"
  git_ssh_key = file("~/.ssh/agentcy-deploy")  # Sensitive
  enabled     = true
}

Sync is RPC, not declarative — invoke POST /api/v1/policies/sources/{id}/sync after apply.

---

agentcy_channel

Messaging channel record (Slack, WhatsApp, Telegram, …). The provider manages the row only — connection (QR scan, OAuth handshake) is completed via the UI/API after the resource is created.

resource "agentcy_channel" "telegram" {
  channel_type  = "telegram"          # Required, immutable
  display_name  = "Acme Bot"
  system_prompt = "Reply concisely."
  source_ids    = [agentcy_source.repo.id]
  response_mode = "agent"             # "disabled", "enabled", or "agent"
  is_active     = true
  filter_config = jsonencode({})      # Optional JSON
}

instance_name is server-derived from channel_type and any existing instances.

---

agentcy_gateway

External orchestration engine connection (OpenFang, future providers). Manages the connection record — does not boot the engine.

resource "agentcy_gateway" "openfang" {
  name        = "primary"
  engine_type = "openfang"
  base_url    = "http://openfang:4200"
  api_key     = var.openfang_key      # Sensitive
  is_default  = true
  context     = jsonencode({})        # Optional JSON passed on every dispatch
}

last_health_status is computed and updates after the API's health-check RPC.