Settings reference

Settings live in the adm_settings table, one row per key, each with a current value, the shipped default, a description and a category. They are edited in the administration section of the application, and readable from PL/SQL.

Reading and writing settings

Read any setting by key:

select adm_settings_api.get_setting_value('CLEAN_TRASH_DAYS') from dual;

Most settings also have a typed getter. Prefer it: it converts, applies the fallback when a row is missing or unparseable, and most of them are result_cached:

begin
  if adm_settings_api.is_object_storage_enabled then ...
  l_days := adm_settings_api.get_clean_trash_days;
end;
/

Admin-only settings

Three settings are marked is_admin_only, and get_setting_value refuses to return them unless the current context is an administrator: HASH_PRIVATE_KEY, AOP_API_KEY and OBJECT_STORAGE_CREDENTIAL_ID (plus AI_QDRANT_WEB_CREDENTIAL_ID and AI_AGENT_WEB_CREDENTIAL_ID in the AI Pack). A read that returns an error rather than a value needs an admin context first — see The security model.

Because most getters are result-cached, a value you change in one session is not necessarily picked up instantly by another. Rule that out first when a change appears to have no effect.

General

Key	Default	What it does
DOWNLOAD_FILE_THRESHOLD	50000000	Size in bytes above which the document page offers a download instead of rendering the file in the viewer. Raise it if your users have the bandwidth and want previews of large files; lower it if they do not.
FILE_METADATA_EXTRACTION	Y	Extract technical metadata (title, author, page count, image dimensions, …) on upload and store it as file.* annotations. See Storage and versioning.
FILE_METADATA_BACKFILL_BATCH_SIZE	500	How many pre-existing documents the daily job backfills metadata for per run.

Maintenance

Key	Default	What it does
CLEAN_TRASH_DAYS	30	How long a trashed document stays in the trash before the daily job deletes it permanently. A retention decision, not a technical one.

Security

Key	Default	What it does
HASH_PRIVATE_KEY	randomly generated at install	The secret used to checksum share-link and embed URL parameters. Admin-only.

Do not change HASH_PRIVATE_KEY casually

Every share link and embed URL already handed out is signed with the current key. Changing it invalidates all of them at once. It is generated per installation at install time, so it is already unique to you — there is no reason to rotate it unless you believe it leaked.

Office services (AOP / AOE)

Key	Default	What it does
AOP_API_URL	https://api.apexofficeprint.com/	APEX Office Print endpoint, used for converting documents to PDF. Point it at your own on-premise AOP server if you run one.
AOP_API_KEY	(empty)	AOP API key. Admin-only.
AOE_API_URL	https://api.apexofficeedit.com/	APEX Office Edit endpoint, used for editing Office documents in the browser.
AOE_API_KEY	(empty)	AOE API key.

→ Editing and converting documents

Object storage

None of these have any effect while OBJECT_STORAGE_ENABLED is N.

Key	Default	What it does
OBJECT_STORAGE_ENABLED	N	Store the content of new files in OCI Object Storage. Turning it on does not move existing files; moving them is a migration.
OBJECT_STORAGE_BUCKET	(empty)	Bucket name.
OBJECT_STORAGE_REGION	(empty)	OCI region identifier, e.g. eu-frankfurt-1.
OBJECT_STORAGE_NAMESPACE	(empty)	OCI object storage namespace.
OBJECT_STORAGE_CREDENTIAL_ID	(empty)	Static ID of the APEX Web Credential holding the OCI credentials. Admin-only.
OBJECT_STORAGE_INDEX	N	Extract text from object-storage files so their content is searchable. Without it, files in the bucket are findable by name only.
OBJECT_STORAGE_VERIFY_CHECKSUMS	Y	Verify the checksum when storing, and re-verify stored objects over time.
CHECKSUM_VERIFICATION_BATCH_SIZE	100	Files whose checksums the daily job re-verifies per run.
CHECKSUM_REVERIFICATION_DAYS	30	How long before a file is due for re-verification again.
MIGRATION_BATCH_SIZE	100	Files moved per migration batch.
OBJECT_STORAGE_RETRY_ATTEMPTS	3	Retries for a failed object storage operation before it is left for the next run.
INDEXING_MAX_ATTEMPTS	3	Text-extraction attempts before a file is marked permanently unindexable and stops being retried. Some files are rejected by the Oracle Text filter every single time; without this budget one of them would fail the daily job forever.

→ Object storage setup · adm_settings_api.configure_object_storage sets the whole group in one call, and test_object_storage_connectivity checks it.

File browser

These apply to the <adm-file-browser> region.

Key	Default	What it does
FS_DEFAULT_VIEW_MODE	TABLE	The view a user who has expressed no preference yet gets: TABLE or GRID.
FS_DEFAULT_PAGE_SIZE	200	Rows fetched per page. Clamped to the component’s own maximum.
FS_UPLOAD_CHUNK_BYTES	786432	Size of one chunk of a chunked upload. Must be a multiple of 3 — otherwise the base64 of a non-final chunk cannot be decoded on its own.
FS_UPLOAD_MAX_FILE_BYTES	2147483648 (2 GiB)	Largest file an upload may announce. Checked before any content is transferred, so an oversized file fails immediately instead of after twenty minutes.
FS_UPLOAD_STAGING_TTL_HOURS	24	How long a partially received upload is kept in adm_upload_staging before the cleanup job discards it.
FS_TIP_FOLDER_ROLLUP	Y	Whether the hover card computes a folder’s recursive size and item counts. The only part of the card that costs real time — a range scan over the folder’s subtree — so turn it off on an instance with a pathological tree. N leaves the rest of the card intact and omits the contents line.
FS_TIP_FOLDER_ROLLUP_MAX_DOCS	50000	How many documents that rollup aggregates before it stops. Past the cap the counts and the size are what was actually reached and the card shows 1.4 GB+ — a partial answer rather than a spinner. A value below 1 is ignored.

Note

The rollup is asked for one folder at a time, and only after the pointer has rested on it — the card does not prefetch, and it does not reuse its warm window for a details card, so every row costs a fresh wait and a pointer travelling down a list issues no requests at all. Folder access in ADM is path-prefix inherited, so a user who may see a folder may see its whole subtree, and the aggregate counts exactly what they could reach by clicking.

Note

On an upgrade these rows are merged rather than re-inserted: the description, category and shipped default are refreshed, but a value you changed is left alone. Raising FS_UPLOAD_MAX_FILE_BYTES is not undone by installing a new release.

AI Pack

Only present once the AI Pack is installed.

Key	Default	What it does
AI_QDRANT_URL	http://localhost:6333	Qdrant endpoint holding the vectors.
AI_QDRANT_COLLECTION_PREFIX	uc_adm_	Prefix for the Qdrant collection names ADM creates, so one Qdrant instance can serve several systems.
AI_QDRANT_WEB_CREDENTIAL_ID	QDRANT_CREDENTIAL	Static ID of the APEX Web Credential holding the Qdrant API key. Admin-only.
AI_CHUNK_FILE_LIMIT	100	Files chunked and embedded per run. The throttle on how hard a sync hits your embedding provider.
AI_TEXT_EXTRACTION	Y	Extract text content into the search index so RAG collections and agent tools can read object-storage files.

The bundled agents’ model

These five decide which provider and model the two shipped agents — the document assistant and the single-document agent — run on. uc_ai speaks to OpenAI, Anthropic, Google, Ollama, OCI, xAI, OpenRouter and Mistral; these settings choose between them.

All five ship empty. While AI_AGENT_PROVIDER is empty, ADM writes nothing and the prompt profiles run on whatever the AI Pack install created them with (OpenAI, at the time of writing) or on whatever you set through the uc_ai APIs yourself. Filling the provider in hands the profiles over to these settings.

Key	Default	What it does
AI_AGENT_PROVIDER	(empty)	openai, anthropic, google, ollama, oci, xai, openrouter, mistral or responses_api. Empty means hands off, as above.
AI_AGENT_MODEL	(empty)	The model id, e.g. gpt-5.4, claude-sonnet-4-6, gemini-3-pro-preview, llama3.1. Required as soon as a provider is set — a provider without a model is refused rather than half-applied.
AI_AGENT_WEB_CREDENTIAL_ID	(empty)	Static ID of the APEX Web Credential holding that provider’s API key. Admin-only. Empty removes the credential from the profile, which is what a local Ollama wants — the credential belongs to the provider you picked, so the previous one is not carried over.
AI_AGENT_REASONING_LEVEL	(empty)	NONE, LOW, MEDIUM or HIGH. Set NONE for a model without reasoning support, which otherwise fails the call outright. Empty leaves the profile’s own reasoning alone.
AI_AGENT_MODEL_CONFIG	(empty)	Optional JSON object merged over everything above, for what the four settings do not reach: {"g_base_url":"http://localhost:11434"} for a self-hosted endpoint, {"anthropic":{"g_max_tokens":8192}} for a provider section. Applied last, so it can also correct a key set above. Everything uc_ai accepts in a model config goes through here — see the UC AI documentation.

Saving applies them

A change to any of these five is written straight into the uc_ai prompt profiles ADM_DOC_ASSISTANT and ADM_DOC_AGENT by a trigger, so it takes effect on the next agent run — from this page, from SQLcl, from a script alike. Only the model-related parts are rewritten: the system prompts, the tool tags (including the one uc_ai_memory added, so agent memory keeps working) and the tool-call limits are left exactly as they are.

A value the provider would reject is refused when you save it rather than at the first agent run, so an unknown provider or a model config that is not JSON comes back as an error on the page.

→ AI agent tools · RAG configurations

Also refer to

• Jobs and maintenance — the batch sizes above only mean something in the context of the job that consumes them
• Administration
• adm_settings_api