feat(ops): add caddy gateway and fix kb dashboard self-host setup

Author: nedasvi

.env.example

@@ -24,6 +24,8 @@ IAA_WS_OWNER_TTL_SECONDS=120
 IAA_WS_OUTBOX_TTL_SECONDS=300
 IAA_WS_TOOL_RESULT_TTL_SECONDS=300
 IAA_KNOWLEDGE_BASES_BASE_URL=http://kb-service:8100
+# Required. Must be a random secret shared with KBS_SERVICE_JWT_SIGNING_KEY.
+# If left as the placeholder value, dashboard KB endpoints will fail.
 IAA_KNOWLEDGE_BASES_SIGNING_KEY=change-me-kb-service-signing-key
 IAA_KNOWLEDGE_BASES_AUDIENCE=kb-service
 IAA_KNOWLEDGE_BASES_JWT_ALGORITHM=HS256
@@ -42,6 +44,7 @@ DATABASE_URL=postgresql://agent:postgres@db:5432/agent
 # Knowledge base service
 KBS_DATABASE_URL=postgresql+asyncpg://agent:postgres@kb-db:5432/knowledge_bases
 KBS_POSTGRES_DB=knowledge_bases
+# Required. Must match IAA_KNOWLEDGE_BASES_SIGNING_KEY.
 KBS_SERVICE_JWT_SIGNING_KEY=change-me-kb-service-signing-key
 KBS_SERVICE_JWT_AUDIENCE=kb-service
 KBS_ENCRYPTION_KEY=change-me-generate-with-python-fernet
@@ -62,3 +65,23 @@ GATEWAY_HTTPS_BIND_ADDR=0.0.0.0
 GATEWAY_HTTPS_BIND_PORT=443
 RESOLVEKIT_PUBLIC_HOST=support.example.com
 LETSENCRYPT_EMAIL=devops@example.com
+
+# Optional: dedicated Dockerized Caddy gateway (`infra/caddy`).
+# This mode supports one main host + explicit `www`/`dash` + `api` hosts,
+# while keeping path routing (`/agent/*`, `/v1/*`) available on the main host.
+CADDY_DOCKER_NETWORK=resolvekit_default
+CADDY_HTTP_BIND=0.0.0.0:80
+CADDY_HTTPS_BIND=0.0.0.0:443
+CADDY_LOCAL_BIND=127.0.0.1:8080
+
+# Public domains served by infra/caddy/Caddyfile
+CADDY_PRIMARY_HOST=support.example.com
+CADDY_WWW_HOST=www.support.example.com
+CADDY_DASH_HOST=dash.support.example.com
+CADDY_API_HOST=api.support.example.com
+
+# Internal upstream targets (override only if your service/container names differ)
+CADDY_DASHBOARD_UPSTREAM=resolvekit_dashboard:3000
+CADDY_API_UPSTREAM=resolvekit_api:3002
+CADDY_BACKEND_UPSTREAM=resolvekit_backend:8000
+CADDY_KB_UPSTREAM=resolvekit_kb_service:8100

README.md

@@ -21,6 +21,12 @@ ResolveKit Backend provides the server-side runtime for embedded app assistants.
    - `cp .env.local-deploy.example .env.local-deploy`
    - configure your public hostname + Let's Encrypt values in `.env.local-deploy`
    - `docker compose -f docker-compose.local-deploy.yml --env-file .env --env-file .env.local-deploy up -d --build`
+5. Optional standalone Dockerized Caddy gateway:
+   - configure `CADDY_*` domain/bind values in `.env`
+   - `docker compose -f infra/caddy/docker-compose.yml up -d`
+6. Optional Caddy from main compose (recommended if you want one command family):
+   - configure `CADDY_*` values in `.env`
+   - `docker compose --profile gateway up -d`
 
 ## Deployment Modes
 

dashboard/next.config.ts

@@ -1,10 +1,31 @@
 import type { NextConfig } from "next";
 
+const allowedDevOrigins = Array.from(
+  new Set(
+    [
+      process.env.CADDY_PRIMARY_HOST,
+      process.env.CADDY_WWW_HOST,
+      process.env.CADDY_DASH_HOST,
+      process.env.CADDY_API_HOST,
+      process.env.RESOLVEKIT_PUBLIC_HOST,
+      process.env.RESOLVEKIT_CONSOLE_HOST,
+      process.env.RESOLVEKIT_API_HOST,
+      process.env.RESOLVEKIT_AGENT_HOST,
+      process.env.NEXT_ALLOWED_DEV_ORIGINS,
+    ]
+      .filter(Boolean)
+      .flatMap((value) => String(value).split(","))
+      .map((value) => value.trim())
+      .filter(Boolean),
+  ),
+);
+
 const nextConfig: NextConfig = {
   reactStrictMode: false,
   // Dashboard and API run separate Next dev processes from the same source tree.
   // Allow overriding distDir so they do not contend on a shared .next cache.
   distDir: process.env.NEXT_DIST_DIR ?? ".next",
+  ...(allowedDevOrigins.length > 0 ? { allowedDevOrigins } : {}),
 };
 
 export default nextConfig;

dashboard/src/app/icon.svg

@@ -5,6 +5,11 @@ import "./globals.css";
 export const metadata: Metadata = {
   title: "ResolveKit Dashboard",
   description: "ResolveKit control plane",
+  icons: {
+    icon: "/icon.svg",
+    shortcut: "/icon.svg",
+    apple: "/icon.svg",
+  },
 };
 
 export default function RootLayout({

dashboard/src/app/layout.tsx

@@ -125,7 +125,17 @@ async function doFetch(path: string, init: RequestInit): Promise<Response> {
 }
 
 async function callInternal(path: string, payload: Record<string, unknown>, ctx: ActorContext): Promise<Record<string, unknown>> {
-  const token = await buildServiceToken(ctx);
+  let token: string;
+  try {
+    token = await buildServiceToken(ctx);
+  } catch (error) {
+    console.error("KB token generation failed", error);
+    throw new KBServiceError({
+      status: 500,
+      detail: "Knowledge base integration is misconfigured",
+      code: "kb_auth_misconfigured",
+    });
+  }
   const response = await doFetch(path, {
     method: "POST",
     headers: {
@@ -155,7 +165,17 @@ export async function callInternalMultipart(
   file: { filename: string; content: Uint8Array; contentType: string },
   ctx: ActorContext,
 ): Promise<Record<string, unknown>> {
-  const token = await buildServiceToken(ctx);
+  let token: string;
+  try {
+    token = await buildServiceToken(ctx);
+  } catch (error) {
+    console.error("KB token generation failed", error);
+    throw new KBServiceError({
+      status: 500,
+      detail: "Knowledge base integration is misconfigured",
+      code: "kb_auth_misconfigured",
+    });
+  }
   const form = new FormData();
   for (const [key, value] of Object.entries(fields)) {
     form.set(key, value);

dashboard/src/lib/server/kb-service.ts

@@ -109,6 +109,7 @@ services:
     working_dir: /app
     depends_on:
       - api
+    env_file: .env
     command: >
       sh -lc "set -e;
       rm -rf .next-dashboard;
@@ -146,6 +147,7 @@ services:
     working_dir: /app
     depends_on:
       - backend
+    env_file: .env
     command: >
       sh -lc "set -e;
       rm -rf .next-api;
@@ -177,6 +179,29 @@ services:
     networks:
       - resolvekit
 
+  caddy:
+    image: caddy:2.8.4
+    container_name: resolvekit_caddy
+    restart: unless-stopped
+    profiles: ["gateway"]
+    depends_on:
+      - backend
+      - kb-service
+      - dashboard
+      - api
+    env_file: .env
+    ports:
+      - "${CADDY_HTTP_BIND:-0.0.0.0:80}:80"
+      - "${CADDY_HTTPS_BIND:-0.0.0.0:443}:443"
+      - "${CADDY_HTTPS_BIND:-0.0.0.0:443}:443/udp"
+      - "${CADDY_LOCAL_BIND:-127.0.0.1:8080}:8080"
+    volumes:
+      - ./infra/caddy/Caddyfile:/etc/caddy/Caddyfile:ro
+      - ./infra/caddy/data:/data
+      - ./infra/caddy/config:/config
+    networks:
+      - resolvekit
+
 networks:
   resolvekit:
     name: resolvekit_default

docker-compose.yml

@@ -9,6 +9,20 @@
 
 ## Local deploy notes
 
+### Optional Dockerized Caddy gateway
+
+If you want Caddy separate from the main compose stack, use `infra/caddy`:
+
+- configure `CADDY_PRIMARY_HOST`, `CADDY_WWW_HOST`, `CADDY_DASH_HOST`, `CADDY_API_HOST`, and `LETSENCRYPT_EMAIL` in `.env`
+- start gateway: `docker compose -f infra/caddy/docker-compose.yml up -d`
+- this gateway reads `.env` directly and proxies to the existing local Docker services on `resolvekit_default`
+
+If you prefer keeping everything in the main compose file, use:
+
+- `docker compose --profile gateway up -d`
+
+This starts the same Caddy config via the `caddy` service in `docker-compose.yml`.
+
 ### Single-host quickstart
 
 The local deploy templates are optimized for one public host:

docs/backend/runbooks/local-dev-and-docker.md

@@ -0,0 +1,62 @@
+{
+	email {$LETSENCRYPT_EMAIL}
+	servers {
+		# Keep original client IP/proto when running behind another private edge.
+		trusted_proxies static private_ranges
+	}
+}
+
+(resolvekit_routes) {
+	encode zstd gzip
+
+	@agent path /agent /agent/*
+	handle @agent {
+		uri strip_prefix /agent
+		reverse_proxy {$CADDY_BACKEND_UPSTREAM:resolvekit_backend:8000} {
+			flush_interval -1
+			transport http {
+				read_timeout 0
+				write_timeout 0
+			}
+		}
+	}
+
+	@kb path /kb /kb/*
+	handle @kb {
+		uri strip_prefix /kb
+		reverse_proxy {$CADDY_KB_UPSTREAM:resolvekit_kb_service:8100}
+	}
+
+	@api_v1 path /v1 /v1/*
+	handle @api_v1 {
+		reverse_proxy {$CADDY_API_UPSTREAM:resolvekit_api:3002}
+	}
+
+	@api_alias path /api /api/*
+	handle @api_alias {
+		uri strip_prefix /api
+		reverse_proxy {$CADDY_API_UPSTREAM:resolvekit_api:3002}
+	}
+
+	handle {
+		reverse_proxy {$CADDY_DASHBOARD_UPSTREAM:resolvekit_dashboard:3000}
+	}
+}
+
+{$CADDY_PRIMARY_HOST:support.example.com}, {$CADDY_WWW_HOST:www.support.example.com}, {$CADDY_DASH_HOST:dash.support.example.com} {
+	import resolvekit_routes
+}
+
+{$CADDY_API_HOST:api.support.example.com} {
+	reverse_proxy {$CADDY_BACKEND_UPSTREAM:resolvekit_backend:8000} {
+		flush_interval -1
+		transport http {
+			read_timeout 0
+			write_timeout 0
+		}
+	}
+}
+
+:8080 {
+	import resolvekit_routes
+}

infra/caddy/Caddyfile

@@ -0,0 +1,34 @@
+# Caddy Gateway (Docker)
+
+This stack runs Caddy as an external reverse proxy for the local OSS Docker services.
+
+## Start
+
+1. Ensure backend stack is up:
+   - `docker compose up -d`
+2. Configure domains in root `.env`:
+   - `CADDY_PRIMARY_HOST`
+   - `CADDY_WWW_HOST`
+   - `CADDY_DASH_HOST`
+   - `CADDY_API_HOST`
+   - `LETSENCRYPT_EMAIL`
+3. Start Caddy:
+   - `docker compose -f infra/caddy/docker-compose.yml up -d`
+
+## Routing
+
+- `https://<CADDY_PRIMARY_HOST>/` -> dashboard (`resolvekit_dashboard:3000`)
+- `https://<CADDY_PRIMARY_HOST>/v1/*` -> dashboard API (`resolvekit_api:3002`)
+- `https://<CADDY_PRIMARY_HOST>/agent/*` -> backend runtime (`resolvekit_backend:8000`)
+- `https://<CADDY_PRIMARY_HOST>/kb/*` -> KB service (`resolvekit_kb_service:8100`)
+- `https://<CADDY_API_HOST>/` -> backend runtime (`resolvekit_backend:8000`)
+
+`/api/*` is also supported as an alias for `/v1/*`.
+
+## Notes
+
+- `CADDY_DOCKER_NETWORK` defaults to `resolvekit_default` (the network from `docker-compose.yml`).
+- If another edge proxy already owns public `:80/:443`, set:
+  - `CADDY_HTTP_BIND=127.0.0.1:18080`
+  - `CADDY_HTTPS_BIND=127.0.0.1:18443`
+- Keep `CADDY_LOCAL_BIND` on loopback for tailnet or local-only access (default `127.0.0.1:8080`).