chore: migrate to single config model (#54)

* chore: migrate inference providers temporarily to native_override to utilize single-file config

Signed-off-by: Jordan Dubrick <jdubrick@redhat.com>

* update lcore image

Signed-off-by: Jordan Dubrick <jdubrick@redhat.com>

* add okp compose file

Signed-off-by: Jordan Dubrick <jdubrick@redhat.com>

* fix numbering

Signed-off-by: Jordan Dubrick <jdubrick@redhat.com>

* add deprecatio notice to llama stack config dir

Signed-off-by: Jordan Dubrick <jdubrick@redhat.com>

* update rag section with new schema

Signed-off-by: Jordan Dubrick <jdubrick@redhat.com>

* set empty default for tokens

Signed-off-by: Jordan Dubrick <jdubrick@redhat.com>

* update lcore image

Signed-off-by: Jordan Dubrick <jdubrick@redhat.com>

* use byo-llm baseline to adopt unified config (single-file)

Signed-off-by: Jordan Dubrick <jdubrick@redhat.com>

* remove llama config and rag content from local compose

Signed-off-by: Jordan Dubrick <jdubrick@redhat.com>

* remove native override for inference

Signed-off-by: Jordan Dubrick <jdubrick@redhat.com>

* add back transformers

Signed-off-by: Jordan Dubrick <jdubrick@redhat.com>

---------

Signed-off-by: Jordan Dubrick <jdubrick@redhat.com>

Author: Jdubrick

.github/workflows/ci-compose-tests.yml

@@ -37,7 +37,7 @@ jobs:
         run: make get-rag CONTAINER_ENGINE=docker
 
       - name: Strip OKP config for smoke test
-        run: yq -i 'del(.rag) | del(.okp)' lightspeed-core-configs/lightspeed-stack.yaml
+        run: yq -i 'del(.rag)' lightspeed-core-configs/lightspeed-stack.yaml
 
       - name: Start services
         run: docker compose --env-file env/default-values.env -f compose/compose.yaml up -d --wait --wait-timeout 120

Makefile

@@ -23,6 +23,7 @@ ENV_FILES += --env-file env/values.env
 endif
 
 LOCAL_COMPOSE_FILES := -f compose/compose.yaml
+OKP_COMPOSE_FILES := -f compose/compose.yaml -f compose/compose-okp.yaml
 
 LIGHTSPEED_STACK_CONFIG := lightspeed-core-configs/lightspeed-stack.yaml
 ifneq ($(wildcard lightspeed-core-configs/lightspeed-stack.local.yaml),)
@@ -48,12 +49,16 @@ get-skills: ## Fetch RHDH skills from GitHub into the skills/ directory
 	bash scripts/fetch-skills.sh
 
 .PHONY: local-up
-local-up:
+local-up: ## Start local compose services
 	$(COMPOSE) $(ENV_FILES) $(LOCAL_COMPOSE_FILES) up -d
 
+.PHONY: local-up-okp
+local-up-okp: ## Start local compose services with OKP
+	$(COMPOSE) $(ENV_FILES) $(OKP_COMPOSE_FILES) up -d
+
 .PHONY: local-down
-local-down:
-	$(COMPOSE) $(ENV_FILES) $(LOCAL_COMPOSE_FILES) down
+local-down: ## Stop local compose services (including OKP if it was started)
+	$(COMPOSE) $(ENV_FILES) $(OKP_COMPOSE_FILES) down --remove-orphans
 
 .PHONY: help
 help: ## Show this help screen

compose/compose-okp.yaml

@@ -0,0 +1,49 @@
+#
+#
+# Copyright Red Hat
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+# Overlay: merged only when `make local-up-okp`.
+# Starts OKP first, then Lightspeed Core.
+services:
+  okp:
+    image: ${OKP_IMAGE:-registry.redhat.io/offline-knowledge-portal/rhokp-rhel9:latest}
+    ports:
+      # Host 8081 so localhost does not clash with LCORE's host 8080.
+      # OKP still listens on 8080 inside its container.
+      - "8081:8080"
+      - "8443:8443"
+    environment:
+      - ACCESS_KEY=${OKP_ACCESS_KEY:-}
+    healthcheck:
+      test:
+        [
+          "CMD-SHELL",
+          "curl -fsS http://127.0.0.1:8080/",
+        ]
+      interval: 5s
+      timeout: 3s
+      retries: 20
+      start_period: 30s
+  lightspeed-core:
+    depends_on:
+      okp:
+        condition: service_healthy
+    extra_hosts:
+      # rhokp_url is also the clickable source link, so it must be a
+      # host-reachable address. This makes `localhost` inside LCORE mean
+      # the host (healthchecks still use 127.0.0.1 and are unaffected).
+      - "localhost:host-gateway"
+    environment:
+      - OKP_SERVICE_URL=http://localhost:8081

compose/compose.yaml

@@ -21,14 +21,15 @@ services:
     volumes:
       - ../${LIGHTSPEED_STACK_CONFIG:-lightspeed-core-configs/lightspeed-stack.yaml}:/app-root/lightspeed-stack.yaml:Z
       - ../lightspeed-core-configs/rhdh-profile.py:/app-root/rhdh-profile.py:Z
-      - ../llama-stack-configs/config.yaml:/app-root/config.yaml:Z
-      - ../rag-content:/rag-content:Z
       - ../skills:/app-root/skills:Z,ro
       # Only used when the vertexai provider is configured. Falls back to a
       # harmless empty placeholder so the mount doesn't fail otherwise.
       - ${GOOGLE_APPLICATION_CREDENTIALS_HOST_PATH:-../env/gcp-credentials-placeholder.json}:/app-root/gcp-credentials.json:Z,ro
     environment:
       - SERVICE_HOST=0.0.0.0
+      - KV_STORE_PATH=${KV_STORE_PATH:-/tmp/kvstore.db}
+      - SQL_STORE_PATH=${SQL_STORE_PATH:-/tmp/sql_store.db}
+      - SQLITE_STORE_DIR=${SQLITE_STORE_DIR:-/tmp/llama-stack-files}
     env_file:
       - ../env/values.env
     healthcheck:

docs/CONTRIBUTING.md

@@ -23,7 +23,7 @@
 
    To configure inference providers without editing the git-tracked `lightspeed-stack.yaml`, copy `lightspeed-core-configs/lightspeed-stack.yaml` to `lightspeed-core-configs/lightspeed-stack.local.yaml` and make your edits there. `make local-up` mounts the `.local.yaml` file automatically when it's present, otherwise it falls back to `lightspeed-stack.yaml`. `lightspeed-stack.local.yaml` is gitignored, so it's safe to leave provider config there permanently.
 
-   The tracked `lightspeed-stack.yaml` contains commented stubs for `vllm`, `openai`, and `vertexai`. Copy it to `lightspeed-stack.local.yaml` and uncomment the provider block(s) you need. For GitOps/production, [scripts/generate-gitops-manifests.sh](../scripts/generate-gitops-manifests.sh) uncomments those three providers and adds production `allowed_models`. Ollama (if needed) is added manually in `.local.yaml` — see [docs/PROVIDERS.md](./PROVIDERS.md).
+   The tracked `lightspeed-stack.yaml` uses the `byo-llm` baseline and contains commented stubs for `vllm`, `openai`, and `vertexai` under `inference.providers`. Copy it to `lightspeed-stack.local.yaml` and uncomment the provider block(s) you need. Providers do not belong under `llama_stack.config.native_override`. For GitOps/production, [scripts/generate-gitops-manifests.sh](../scripts/generate-gitops-manifests.sh) uncomments those three providers and adds production `allowed_models`. Ollama (if needed) is added manually in `.local.yaml` — see [docs/PROVIDERS.md](./PROVIDERS.md).
 
 2. Pull the RAG content:
 
@@ -39,26 +39,33 @@ make get-skills
 
 4. The production config (`lightspeed-stack.yaml`) sets `host: 127.0.0.1` so the service only binds to loopback — reachable exclusively by containers in the same Pod on Kubernetes. The compose file overrides this with `SERVICE_HOST=0.0.0.0` so the container port mapping works and you can reach the API at `localhost:8080` from your host.
 
-1. Start the local API stack:
+5. Start the local API stack:
 
 ```sh
 make local-up
 ```
 
-This starts Lightspeed Core using the mounted config/content below.
+This starts Lightspeed Core using the mounted config/content below. To also start the Offline Knowledge Portal (OKP) and wait until it is healthy before Lightspeed Core starts:
+
+```sh
+make local-up-okp
+```
+
+Set `OKP_ACCESS_KEY` in `env/values.env` when using `make local-up-okp`. `make local-down` stops OKP even if you started it with the overlay.
+
+To obtain `OKP_ACCESS_KEY` navigate to the [access key generator](https://access.redhat.com/offline/access/).
 
 Lightspeed Core uses mounted config/content in local compose:
 
 - `lightspeed-core-configs/lightspeed-stack.yaml` (or `lightspeed-stack.local.yaml`, if present) -> `/app-root/lightspeed-stack.yaml`
 - `lightspeed-core-configs/rhdh-profile.py` -> `/app-root/rhdh-profile.py`
-- `llama-stack-configs/config.yaml` -> `/app-root/config.yaml`
 - `rag-content/` -> `/rag-content`
 
 Question validation is not enabled automatically. If you want it, set `ENABLE_VALIDATION=question_validity`, `VALIDATION_PROVIDER`, and `VALIDATION_MODEL_NAME` in `env/values.env`, along with any env vars required by the selected inference provider.
 
 See [Configuring Validation](#configuring-validation) for example configurations.
 
-4. Stop services:
+6. Stop services:
 
 ```sh
 make local-down
@@ -107,7 +114,7 @@ Paste that value into `byok_rag[].vector_db_id`. Keep `embedding_model` as the d
 
 `notebooks` is separate: it is dynamic create capacity under `vector_store` (local FAISS; GitOps rewrites it to pgvector). It is not a second `byok_rag` corpus.
 
-If you use a gitignored `lightspeed-stack.local.yaml` for local providers, copy the same `byok_rag` / `rag` / `vector_store` / `shields` sections from the committed file when they change.
+If you use a gitignored `lightspeed-stack.local.yaml`, copy the same `byok_rag` / `rag` / `vector_store` / `shields` sections from the committed file when they change.
 
 ## Configuring Skills
 
@@ -204,7 +211,8 @@ make validate-yaml
 | `get-rag` | Pull and unpack RAG content into `./rag-content` (replaces existing contents). Optional: `RAG_CONTENT_IMAGE=<image>`. |
 | `get-skills` | Optional. Fetch RHDH skills into `./skills` for skills consumption (replaces existing contents). Optional: `RHDH_SKILLS_REPO=<url>`, `RHDH_SKILLS_REF=<ref>`. |
 | `local-up` | Start local compose services. Validation is controlled entirely through env vars in `env/values.env`. |
-| `local-down` | Stop local compose services. |
+| `local-up-okp` | Start local compose services with OKP (`compose/compose-okp.yaml`). Set `OKP_ACCESS_KEY` in `env/values.env`. |
+| `local-down` | Stop local compose services, including OKP if it was started. |
 | `sync-images` | Sync image values from `images.yaml` into `env/default-values.env`. Requires `yq`. |
 | `validate-images` | Validate that `images.yaml` and `env/default-values.env` are in sync. Requires `yq`. |
 | `validate-yaml` | Validate YAML formatting/syntax. |

docs/PROVIDERS.md

@@ -12,7 +12,7 @@ Each inference has its own environment variables. You can include all of these i
 > You will notice the `api_key_env` field is not wrapped in curly-braces `{}`. This is due to Lightspeed Core wrapping them internally to curate a proper `{env.xyz}` to pass through to Llama Stack so the keys are not exposed internally.
 
 > [!NOTE]
-> Commented provider stubs for `vllm`, `openai`, and `vertexai` live in [lightspeed-stack.yaml](../lightspeed-core-configs/lightspeed-stack.yaml). For local development, copy that file to `lightspeed-core-configs/lightspeed-stack.local.yaml` (gitignored, auto-mounted by `make local-up` when present — see [CONTRIBUTING.md](./CONTRIBUTING.md)), uncomment the block(s) you need, and set the required env vars.
+> Commented provider stubs for `vllm`, `openai`, and `vertexai` live under `inference.providers` in [lightspeed-stack.yaml](../lightspeed-core-configs/lightspeed-stack.yaml). The `byo-llm` baseline means these stay at that top-level section — do not add them to `llama_stack.config.native_override`. For local development, copy that file to `lightspeed-core-configs/lightspeed-stack.local.yaml` (gitignored, auto-mounted by `make local-up` when present — see [CONTRIBUTING.md](./CONTRIBUTING.md)), uncomment the block(s) you need, and set the required env vars.
 >
 > For GitOps/production, [scripts/generate-gitops-manifests.sh](../scripts/generate-gitops-manifests.sh) uncomments those three providers, then adds production `allowed_models` for `openai` and `vertexai`.
 >
@@ -143,10 +143,9 @@ service:
 llama_stack:
   use_as_library_client: true
   config:
-    profile: /app-root/config.yaml
+    baseline: byo-llm
 inference:
   providers:
-    - type: sentence_transformers
     - type: openai
       id: openai
       api_key_env: OPENAI_API_KEY
@@ -184,4 +183,4 @@ mcp_servers:
     url: 'http://localhost:7007/api/mcp-actions/v1'
     authorization_headers:
       Authorization: 'client'
-```
+```

env/default-values.env

@@ -2,9 +2,14 @@
 # You do not need to set them all if they will go unused.
 
 # Service Images
-LIGHTSPEED_CORE_IMAGE=quay.io/lightspeed-core/lightspeed-stack:dev-20260811-0da4b12
+LIGHTSPEED_CORE_IMAGE=quay.io/lightspeed-core/lightspeed-stack:dev-20260824-cbd182b
 RAG_CONTENT_IMAGE=quay.io/redhat-ai-dev/rag-content:release-1.10-lls-0.5.0-8c231a3b5177f12fff9db042dfa4091d8f2f26b3
 
+# Llama Stack storage (default_run.yaml). Matches llama-stack-configs/config.yaml.
+KV_STORE_PATH=/tmp/kvstore.db
+SQL_STORE_PATH=/tmp/sql_store.db
+SQLITE_STORE_DIR=/tmp/llama-stack-files
+
 # vLLM Inference Settings
 VLLM_URL=
 VLLM_API_KEY=
@@ -37,6 +42,9 @@ ENABLE_VALIDATION=
 VALIDATION_PROVIDER=
 VALIDATION_MODEL_NAME=
 
+# Offline Knowledge Portal (used by `make local-up-okp`)
+OKP_ACCESS_KEY=
+
 # Other
 LLAMA_STACK_LOGGING=
 

images.yaml

@@ -15,7 +15,7 @@
 # limitations under the License.
 lightspeed-core:
   llama_version: 0.6.0
-  image: quay.io/lightspeed-core/lightspeed-stack:dev-20260811-0da4b12
+  image: quay.io/lightspeed-core/lightspeed-stack:dev-20260824-cbd182b
   latest_release: v0.6.0rc2 # using v prefix in catalog now it seems
 rag-content:
   llama_version: 0.5.0

lightspeed-core-configs/lightspeed-stack.yaml

@@ -24,7 +24,7 @@ service:
 llama_stack:
   use_as_library_client: true
   config:
-    profile: /app-root/config.yaml
+    baseline: byo-llm
     native_override:
       vector_stores:
         annotation_prompt_params:
@@ -147,9 +147,11 @@ mcp_servers:
     authorization_headers:
       Authorization: 'client'
 rag:
-  tool:
-    - okp
-okp:
-  rhokp_url: '${env.OKP_SERVICE_URL:=http://localhost:8080}'
-  offline: true
-  chunk_filter_query: 'product:*developer_hub*'
+  okp:
+    rhokp_url: '${env.OKP_SERVICE_URL:=http://localhost:8080}'
+    offline: true
+    chunk_filter_query: 'product:*developer_hub*'
+  retrieval:
+    tool:
+      sources:
+        - okp

llama-stack-configs/README.md

@@ -1,5 +1,9 @@
 # Llama Stack Configuration Files
 
+> [!IMPORTANT]
+> This directory is set to be deleted/archived in favour of a lightspeed-stack.yaml only implementation
+>
+
 This directory stores the Llama Stack config for the single active release tracked on `main`.
 
 - `config.yaml` is the unified config. The question-validation shield is conditionally enabled via `ENABLE_VALIDATION` and uses the `VALIDATION_PROVIDER` and `VALIDATION_MODEL_NAME` env vars you supply (see [CONTRIBUTING.md](../docs/CONTRIBUTING.md#configuring-validation)).