first commit

terraform/examples/gcp-gke/README.md

@@ -0,0 +1,37 @@
+# Bifrost on GCP GKE
+
+Deploys Bifrost as a Kubernetes workload on Google Kubernetes Engine.
+
+## Prerequisites
+
+- GCP project with billing enabled
+- `gcloud` CLI authenticated (`gcloud auth application-default login`)
+- Terraform >= 1.0
+- GKE API enabled (`gcloud services enable container.googleapis.com`)
+
+## Usage
+
+```bash
+# Copy and edit the example variables file
+cp terraform.tfvars.example terraform.tfvars
+
+# Deploy
+terraform init
+terraform plan
+terraform apply
+```
+
+## Configuration
+
+Two approaches can be combined:
+
+1. **File-based** -- Set `config_json_file` to point to an existing `config.json`.
+2. **Variable-based** -- Set individual variables (`config_store`, `logs_store`, `providers_config`). These override matching keys from the file.
+
+See `terraform.tfvars.example` for examples of both.
+
+## Cleanup
+
+```bash
+terraform destroy
+```

terraform/examples/gcp-gke/main.tf

@@ -0,0 +1,49 @@
+terraform {
+  required_version = ">= 1.0"
+  required_providers {
+    google = {
+      source  = "hashicorp/google"
+      version = "~> 5.0"
+    }
+    kubernetes = {
+      source  = "hashicorp/kubernetes"
+      version = "~> 2.0"
+    }
+  }
+}
+
+provider "google" {
+  project = var.project_id
+  region  = var.region
+}
+
+module "bifrost" {
+  source         = "../../modules/bifrost"
+  cloud_provider = "gcp"
+  service        = "gke"
+  region         = var.region
+  gcp_project_id = var.project_id
+  image_tag      = var.image_tag
+  name_prefix    = var.name_prefix
+
+  # Config: use a file as base, override with variables
+  config_json_file = var.config_json_file
+
+  # Override specific config sections
+  config_store     = var.config_store
+  logs_store       = var.logs_store
+  providers_config = var.providers_config
+
+  # Compute
+  desired_count        = var.desired_count
+  cpu                  = var.cpu
+  memory               = var.memory
+  create_cluster       = var.create_cluster
+  node_count           = var.node_count
+  create_load_balancer = var.create_load_balancer
+
+  # Autoscaling
+  enable_autoscaling = var.enable_autoscaling
+  min_capacity       = var.min_capacity
+  max_capacity       = var.max_capacity
+}

terraform/examples/gcp-gke/outputs.tf

@@ -0,0 +1,9 @@
+output "service_url" {
+  description = "URL to access the Bifrost service."
+  value       = module.bifrost.service_url
+}
+
+output "health_check_url" {
+  description = "URL to the Bifrost health check endpoint."
+  value       = module.bifrost.health_check_url
+}

terraform/examples/gcp-gke/terraform.tfvars.example

@@ -0,0 +1,66 @@
+# =============================================================================
+# GCP GKE Example — terraform.tfvars
+# WARNING: Do NOT commit this file with real secrets (API keys, passwords).
+#          Use environment variables, a secrets manager, or .gitignore this file.
+# =============================================================================
+
+project_id  = "my-gcp-project"
+region      = "us-central1"
+image_tag   = "latest"
+name_prefix = "bifrost"
+
+# -----------------------------------------------------------------------------
+# Config approach 1: File-based
+# Point to an existing config.json. Variable overrides below will merge on top.
+# -----------------------------------------------------------------------------
+# config_json_file = "./config.json"
+
+# -----------------------------------------------------------------------------
+# Config approach 2: Variable-based
+# Define config sections directly. These override matching keys from the file.
+# -----------------------------------------------------------------------------
+config_store = {
+  enabled = true
+  type    = "sqlite"
+  config = {
+    path = "/app/data/bifrost.db"
+  }
+}
+
+logs_store = {
+  enabled = true
+  type    = "sqlite"
+  config = {
+    path = "/app/data/bifrost-logs.db"
+  }
+}
+
+providers_config = {
+  openai = {
+    api_key = "sk-..."
+  }
+  vertex = {
+    project_id = "my-gcp-project"
+    region     = "us-central1"
+  }
+}
+
+# -----------------------------------------------------------------------------
+# Compute
+# -----------------------------------------------------------------------------
+desired_count        = 2
+cpu                  = 500
+memory               = 1024
+create_cluster       = true
+node_count           = 3
+create_load_balancer = true
+
+# -----------------------------------------------------------------------------
+# Autoscaling
+# -----------------------------------------------------------------------------
+# NOTE: If you are using OSS version - running multiple nodes has an effect on
+# functionality of the system. Please read
+# https://docs.getbifrost.ai/deployment-guides/how-to/multinode
+enable_autoscaling = false
+# min_capacity       = 1
+# max_capacity       = 5

terraform/examples/gcp-gke/variables.tf

@@ -0,0 +1,110 @@
+variable "project_id" {
+  description = "GCP project ID."
+  type        = string
+}
+
+variable "region" {
+  description = "GCP region to deploy into."
+  type        = string
+  default     = "us-central1"
+}
+
+variable "image_tag" {
+  description = "Bifrost Docker image tag."
+  type        = string
+  default     = "latest"
+}
+
+variable "name_prefix" {
+  description = "Prefix for all resource names."
+  type        = string
+  default     = "bifrost"
+}
+
+# --- Config: file-based ---
+
+variable "config_json_file" {
+  description = "Path to a Bifrost config.json file. Variables below override matching keys."
+  type        = string
+  default     = null
+}
+
+# --- Config: variable-based overrides ---
+
+variable "config_store" {
+  description = "Config store configuration (type: sqlite/postgres)."
+  type        = any
+  default     = null
+}
+
+variable "logs_store" {
+  description = "Logs store configuration (type: sqlite/postgres)."
+  type        = any
+  default     = null
+}
+
+variable "providers_config" {
+  description = "LLM provider configurations (openai, anthropic, vertex, etc.)."
+  type        = any
+  default     = null
+}
+
+# --- Compute ---
+
+variable "desired_count" {
+  description = "Number of Kubernetes pods."
+  type        = number
+  default     = 1
+}
+
+variable "cpu" {
+  description = "CPU in millicores for each pod."
+  type        = number
+  default     = 500
+}
+
+variable "memory" {
+  description = "Memory in MB for each pod."
+  type        = number
+  default     = 1024
+}
+
+# --- Cluster ---
+
+variable "create_cluster" {
+  description = "Create a new GKE cluster. Set to false to use an existing cluster."
+  type        = bool
+  default     = true
+}
+
+variable "node_count" {
+  description = "Number of nodes in the GKE node pool."
+  type        = number
+  default     = 3
+}
+
+variable "create_load_balancer" {
+  description = "Create a load balancer via Kubernetes Ingress."
+  type        = bool
+  default     = true
+}
+
+# --- Autoscaling ---
+
+variable "enable_autoscaling" {
+  description = "Enable Horizontal Pod Autoscaler."
+  type        = bool
+  default     = false
+}
+
+variable "min_capacity" {
+  description = "Minimum number of pods when autoscaling is enabled."
+  type        = number
+  default     = 1
+}
+
+variable "max_capacity" {
+  description = "Maximum number of pods when autoscaling is enabled."
+  type        = number
+  default     = 10
+}