Removed unused config values - llm info should only be set in plugin config.

plugin/src/client.ts

@@ -135,11 +135,17 @@ export const runPipeline = async ({
         }),
     })
 
-    const data = await readApiResponse<PipelineRunResponse | { detail?: string }>(response, url)
+    const data = await readApiResponse<PipelineRunResponse | { detail?: unknown }>(response, url)
 
     if (!response.ok) {
+        if (response.status == 422) {
+            throw new Error(
+                'Plugin is not configured. Set the model, API key, and API URL in the plugin settings.',
+            )
+        }
+        const detail = 'detail' in data ? data.detail : undefined
         throw new Error(
-            'detail' in data && data.detail ? data.detail : 'Pipeline execution failed.',
+            typeof detail === 'string' && detail ? detail : 'Pipeline execution failed.',
         )
     }
 

service/README.md

@@ -13,28 +13,6 @@ Create `config.yaml` with your LLM credentials (see `config.template.yaml`). All
 prompts file, can be configured there. To use a different config file, set `AI_DOCUMENT_PLUGIN_CONFIG_PATH` to its path
 before starting the API or CLI.
 
-Run the full pipeline:
-
-```bash
-uv run ai-document-pipeline \
-  --questionnaire-uuid <QUESTIONNAIRE_UUID> \
-  --token <TOKEN> \
-  --template-uuid <TEMPLATE_UUID>
-```
-
-The CLI reads technical configuration such as DSW API base URL, database connection, model configuration, prompt paths,
-and output file paths from `config.yaml` or from the file referenced by `AI_DOCUMENT_PLUGIN_CONFIG_PATH`. The runtime
-inputs `questionnaire_uuid`, `token`, and `template_uuid` are passed explicitly through CLI arguments.
-
-This produces:
-
-| File                       | Description                                      |
-|----------------------------|--------------------------------------------------|
-| `dmp_output_pre_polish.md` | Generated DMP with debug tables before polishing |
-| `dmp_output.md`            | Final polished DMP with token-usage summary      |
-
-Step 1 produces new record in the `assignments` table in the DB, when questions are assigned for the first time.
-
 ## Pipeline steps
 
 `src/ai_document_plugin_service/run_pipeline.py` orchestrates three steps, each backed by a dedicated module:
@@ -55,12 +33,12 @@ consolidating duplicates, and improving flow — without adding new content.
 
 ## Configuration
 
+The LLM API information (url, api key, model) must be set in the plugin configuration in the deployed DSW instance
+(Administration tab > Settings > Plugins).
+
 ### `config.yaml`
 
-LLM connection settings (API key, base URL, model name), DSW API base URL, database connection, and all pipeline file
-paths. The API key supports environment variable expansion (e.g. `$OPENAI_API_KEY`). The file itself can be overridden
-globally via `AI_DOCUMENT_PLUGIN_CONFIG_PATH`. The config file does not contain its own `config_path`; the active file
-is determined only by runtime startup/defaults.
+The file path can be overridden globally via `AI_DOCUMENT_PLUGIN_CONFIG_PATH`. See `config.template.yaml` for more info.
 
 ### `prompts.yaml`
 
@@ -79,7 +57,8 @@ The project `Makefile` provides a few shortcuts for common development tasks:
 - `make dev` starts the FastAPI development server with auto-reload on port `8010`
 - `make build` builds the Python package
 - `make db` starts the local PostgreSQL container defined in `docker-compose.yml`
-- `make db-init` starts the local PostgreSQL container, waits for it to become healthy, and applies all Alembic migrations
+- `make db-init` starts the local PostgreSQL container, waits for it to become healthy, and applies all Alembic
+  migrations
 - `make db-migrate` applies all Alembic migrations to the configured database
 - `make db-current` shows the current Alembic revision stored in the database
 - `make db-history` shows available Alembic migration history
@@ -95,13 +74,18 @@ make dev
 ```
 
 `docker compose` creates the PostgreSQL database itself from the `POSTGRES_DB`, `POSTGRES_USER`, and
-`POSTGRES_PASSWORD` values in [docker-compose.yml](/Users/hana/DSW/AI-playground/ai-document-plugin/service/docker-compose.yml:1). On application startup, the service verifies that it can connect to the configured database and runs Alembic migrations to `head`. Alembic creates or updates the schema inside that database; it does not create the PostgreSQL server or database on its own.
+`POSTGRES_PASSWORD` values
+in [docker-compose.yml](/Users/hana/DSW/AI-playground/ai-document-plugin/service/docker-compose.yml:1). On application
+startup, the service verifies that it can connect to the configured database and runs Alembic migrations to `head`.
+Alembic creates or updates the schema inside that database; it does not create the PostgreSQL server or database on its
+own.
 
 `make db-init` and `make db-migrate` remain useful local shortcuts when you want to manage migrations manually, but they
 are not required by deployed service startup.
 
-If the PostgreSQL container already exists with an older Docker volume, the database may be missing even though the server is running. In that case `make db-init` also checks that `ai_document_plugin` exists and creates it before running Alembic migrations.
-
+If the PostgreSQL container already exists with an older Docker volume, the database may be missing even though the
+server is running. In that case `make db-init` also checks that `ai_document_plugin` exists and creates it before
+running Alembic migrations.
 
 ## Tests
 

service/config.template.yaml

@@ -1,20 +1,14 @@
-llm_response_generation:
-  type: "openai-api"
-  model: "gpt-oss-120b"
-  api_url: https://llm.ai.e-infra.cz/v1/
-  api_key: "xxx"
-  workers: 4
-
-dsw:
-  api_url: ""
-
+# Specify urls which can be used for authenticating the user. The user will send the url in a header 'X-Dsw-Api-Url'
+# Use the base url for dsw api, not the frontend.
+# For example: https://researchers.dsw.elixir-europe.org/wizard-api
 auth:
   allowed_project_urls:
-    - "https://your-dsw-instance.example.com"
+    - "https://your-dsw-instance.example.com/wizard-api"
 
 logging:
   level: "INFO"
 
+# Specify login details for Postgresql database.
 database:
   host: "localhost"
   port: "5432"
@@ -23,5 +17,6 @@ database:
   password: "ai_document_plugin"
   schema: "public"
 
+# Path to prompts config. Path is set as the relative path from this config.
 files:
   prompts_path: "prompts.yaml"

service/src/ai_document_plugin_service/ai/assignment/assignment_component.py

@@ -24,6 +24,7 @@
 )
 from ai_document_plugin_service.ai.assignment.types import SectionAssignment
 from ai_document_plugin_service.ai.common.config import Config
+from ai_document_plugin_service.ai.common.llm_client import LLMClient
 from ai_document_plugin_service.ai.common.progress import progress_percent
 from ai_document_plugin_service.ai.common.types import AssignmentStats
 from ai_document_plugin_service.ai.knowledgemodel.types import QuestionData
@@ -38,6 +39,17 @@ class AssignmentComponentResult(TypedDict):
 
 @component
 class AssignmentComponent:
+    """
+    This component is responsible for assigning questions from the questionnaire to the sections from the dmp template.
+    For example, it assigns question 'When will the project start?' to sections Introduction and Project Timeline
+    """
+
+    def __init__(self, llm_client: LLMClient, config: Config) -> None:
+        self.llm_client = llm_client
+        self.config = config
+        self.section_id_generator = OpenAISectionIdGenerator(llm_client, config)
+        self.section_matcher = OpenAILayerMatcher(self.llm_client, self.config)
+
     @staticmethod
     def _add_chunk_mapping_to_result(
         *,
@@ -53,16 +65,13 @@ def _add_chunk_mapping_to_result(
                 continue
             result_mapping[question_path] = [section_formatter.record_id_for_sid(sid) for sid in section_ids]
 
-    @staticmethod
     def _match_single_chunk(
-        *,
-        config: Config,
+        self,
         sections_xml: str,
         question_chunk: str,
         stats: AssignmentStats,
     ) -> dict[str, list[str]]:
-        matcher = OpenAILayerMatcher(config)
-        return matcher.match_questions_to_sections(
+        return self.section_matcher.match_questions_to_sections(
             sections_xml,
             question_chunk,
             stats,
@@ -73,7 +82,6 @@ def run(
         self,
         data: list[QuestionData],
         template_data: dict[str, Any],
-        config: Config,
         km: dict[str, Any],
         on_progress: Callable[[str], None] | None = None,
     ) -> AssignmentComponentResult:
@@ -85,7 +93,7 @@ def run(
         stats = AssignmentStats()
 
         section_formatter = SectionFormatter(sections)
-        section_formatter.create_mappings(OpenAISectionIdGenerator(config), stats)
+        section_formatter.create_mappings(self.section_id_generator, stats)
         sections_xml = section_formatter.get_sections_as_xml()
 
         result_mapping = {}
@@ -94,7 +102,6 @@ def run(
 
         def match_chunk(question_chunk: str) -> dict[str, list[str]]:
             result = self._match_single_chunk(
-                config=config,
                 sections_xml=sections_xml,
                 question_chunk=question_chunk,
                 stats=stats,
@@ -109,8 +116,8 @@ def match_chunk(question_chunk: str) -> dict[str, list[str]]:
         for question_to_section_ids in thread_map(
             match_chunk,
             question_chunks,
-            max_workers=config.parallel_workers,
-            desc=f'Assigning questions to sections ({config.parallel_workers} workers)',
+            max_workers=self.llm_client.get_max_workers(),
+            desc=f'Assigning questions to sections ({self.llm_client.get_max_workers()} workers)',
         ):
             self._add_chunk_mapping_to_result(
                 result_mapping=result_mapping,

service/src/ai_document_plugin_service/ai/assignment/llm.py

@@ -45,9 +45,9 @@ def match_questions_to_sections(
 
 
 class OpenAILayerMatcher(LayerMatcher):
-    def __init__(self, config: Config) -> None:
+    def __init__(self, llm_client: LLMClient, config: Config) -> None:
+        self.client = llm_client
         self.config = config
-        self.client = LLMClient(config)
 
     def match_questions_to_sections(
         self,
@@ -74,7 +74,6 @@ def match_questions_to_sections(
 
         def call_and_parse() -> dict[str, list[str]]:
             response = self.client.completion(
-                model=self.config.model,
                 messages=messages,
                 temperature=self.config.assignment.temperature,
                 max_tokens=self.config.assignment.max_tokens,

service/src/ai_document_plugin_service/ai/assignment/sections/llm.py

@@ -1,7 +1,6 @@
 import typing
 from abc import ABC, abstractmethod
 
-from openai import OpenAI
 from tqdm import tqdm
 
 from ai_document_plugin_service.ai.assignment.types import LeafSection
@@ -29,9 +28,9 @@ def generate_leaf_section_ids(
 
 
 class OpenAISectionIdGenerator(SectionIdGenerator):
-    def __init__(self, config: Config) -> None:
+    def __init__(self, llm_client: LLMClient, config: Config) -> None:
         self.config = config
-        self.client = LLMClient(config)
+        self.client = llm_client
 
     @typing.override
     def generate_leaf_section_ids(
@@ -59,7 +58,6 @@ def generate_leaf_section_ids(
             )
             response = call_with_retry(
                 lambda um=user_msg: self.client.completion(
-                    model=self.config.model,
                     messages=[
                         {'role': 'system', 'content': system_msg},
                         {'role': 'user', 'content': um},
@@ -92,7 +90,6 @@ def _normalize_section_id(raw: str) -> str:
 class LoggingNoopSectionIdGenerator(SectionIdGenerator):
     def __init__(self, config: Config) -> None:
         self.config = config
-        self.client = OpenAI(api_key=config.api_key, base_url=config.api_url)
 
     @typing.override
     def generate_leaf_section_ids(  # ty: ignore[invalid-method-override]

service/src/ai_document_plugin_service/ai/common/config.py

@@ -1,6 +1,6 @@
 import os
 import pathlib
-from dataclasses import dataclass, replace
+from dataclasses import dataclass
 from typing import Any
 
 import yaml
@@ -41,26 +41,21 @@ class DatabaseConfig:
 
 @dataclass(frozen=True)
 class Config:
-    api_key: str
-    api_url: str
-    dsw_api_url: str
     allowed_project_urls: tuple[str, ...]
-    model: str
     log_level: str
     database: DatabaseConfig
     files: FilePaths
     assignment: SystemAndUserPrompt
     section_id: SystemAndUserPrompt
     dmp_generation: SystemPrompt
     dmp_polishing: SystemAndUserPrompt
-    parallel_workers: int
 
 
 @dataclass(frozen=True)
-class LLMConfigOverride:
-    model: str | None = None
-    api_key: str | None = None
-    api_url: str | None = None
+class LLMConfig:
+    model: str
+    api_key: str
+    api_url: str
     parallel_workers: int | None = None
 
 
@@ -134,19 +129,6 @@ def _get_allowed_project_urls(config: dict[str, Any]) -> tuple[str, ...]:
     return tuple(normalized)
 
 
-def _get_parallel_workers(config: dict[str, Any]) -> int:
-    workers = config.get('llm_response_generation', {}).get('workers', 1)
-    try:
-        workers_int = int(workers)
-    except (TypeError, ValueError) as exc:
-        msg = "Invalid config value: 'parallelism.workers' must be an integer >= 1"
-        raise ValueError(msg) from exc
-    if workers_int < 1:
-        msg = "Invalid config value: 'parallelism.workers' must be >= 1"
-        raise ValueError(msg)
-    return workers_int
-
-
 def _resolve_existing_path(path: str, *, base_dir: pathlib.Path | None = None) -> str:
     normalized_path = _normalize_path(path)
     path_obj = pathlib.Path(normalized_path)
@@ -199,12 +181,6 @@ def load_config(config_path: str | None = None) -> Config:
         raise TypeError(msg)
 
     return Config(
-        api_key=_expand_env_vars(
-            _get(config, 'llm_response_generation', 'api_key'),
-        ),
-        api_url=_get(config, 'llm_response_generation', 'api_url'),
-        model=_get(config, 'llm_response_generation', 'model'),
-        dsw_api_url=_get(config, 'dsw', 'api_url'),
         allowed_project_urls=_get_allowed_project_urls(config),
         log_level=_get_log_level(config),
         database=DatabaseConfig(
@@ -241,18 +217,4 @@ def load_config(config_path: str | None = None) -> Config:
             system_message=_get(prompts, 'dmp_polishing', 'system_message'),
             user_message=_get(prompts, 'dmp_polishing', 'user_message'),
         ),
-        parallel_workers=_get_parallel_workers(config),
-    )
-
-
-def apply_llm_override(config: Config, override: LLMConfigOverride | None = None) -> Config:
-    if override is None:
-        return config
-
-    return replace(
-        config,
-        model=override.model or config.model,
-        api_key=override.api_key or config.api_key,
-        api_url=override.api_url or config.api_url,
-        parallel_workers=override.parallel_workers or config.parallel_workers,
     )