Infrastructure as Code

Manage your Agentcy platform declaratively. Define agents, connectors, channels, policies, and the rest of your workspace in code, version it in Git, and apply changes the same way you'd apply infrastructure changes anywhere else.

The Terraform provider is the supported IaC interface today. It talks to the same /api/v1/ endpoints the UI does, so anything you can do through the UI you can do through Terraform — and the same RBAC rules apply.

What you can manage

14 resources are e2e-verified and ship in v1:

Resource	What it controls
agentcy_source	Ingestion connectors (GitHub, AWS, GCP, CSV, …)
agentcy_settings	Org-wide LLM and embedding configuration
agentcy_realm	Graph namespace partitions
agentcy_agent	Typed agents (enrichment, anomaly, relationship discovery)
agentcy_webhook	Inbound webhook channels
agentcy_skill	Versioned skill blobs
agentcy_memory	Persistent memory entries
agentcy_api_key	Programmatic API keys
agentcy_task	Cron-triggered scheduled tasks
agentcy_role	RBAC roles
agentcy_policy	Rego policies (with plan-time validation)
agentcy_policy_source	Git-backed policy sources
agentcy_channel	Messaging channels (Slack, WhatsApp, Telegram, …)
agentcy_gateway	External orchestration engine connections

8 read-only data sources for referencing existing state:

Data source	What it reads
agentcy_agent_types	Catalog of valid agent_type values
agentcy_permissions	RBAC permission catalog
agentcy_realm_stats	Live node/relationship counts per realm
agentcy_running_tasks	Tasks currently executing
agentcy_memory_search	Semantic search across memory
agentcy_artifacts	Artifact registry
agentcy_conversations	Chat conversation metadata
agentcy_pipeline_runs	Ingestion pipeline run history

When to use Terraform vs UI vs API

• Terraform: anything you want versioned, reviewed, replicated across environments, or recreated from scratch — agents, connectors, policies, scheduled tasks, RBAC.
• UI: ad-hoc exploration, one-off chats, message replies, uploading artifacts, completing OAuth/QR flows for channels.
• API directly: scripts and one-shot RPCs (run an agent now, sync a source now, send a message). The provider intentionally doesn't model RPC actions — those aren't declarative state.

Quickstart

1. Install the provider
2. Configure it against your tenant
3. Start with the end-to-end example — one HCL file that creates a source, agent, channel, role, and policy, all wired together.

Versioning and releases

The provider is v0.1 — the resource set above is stable, but expect new resources, new attributes on existing resources, and minor schema tweaks before v1.0. Schema changes are flagged in the CHANGELOG. Pin a specific version in production.

What's not in v1

• agentcy_workflow, agentcy_trigger, agentcy_template — these proxy to OpenFang. Coming once OpenFang's image is publicly accessible (the design and code path are already in place).
• agentcy_cloud_instance, agentcy_cloud_team_member — for provisioning new tenants on Agentcy Cloud. Coming when the cloud control plane API is publicly documented.
• File uploads (agentcy_artifact as a managed resource) — binary I/O in HCL is awkward; use the API and reference uploaded artifacts via the agentcy_artifacts data source instead.
• Conversation/message resources — ephemeral state, not declarative.