Atlan

Create one or more glossary terms in Atlan. Use mode='draft' when the user asked to draft, preview, or review the change before it is applied.

create_glossary_terms

Create a new knowledge folder in Atlan. Idempotent — returns the existing folder when it already exists. Knowledge folders organise uploaded documents (PDFs, Markdown, YAML) that agents read for business context.

create_knowledge_folder_tool

Create one SQL-intelligence asset in Atlan so every future semantic-model build and every consumer reuses it: a join (kind=join), a named filter (kind=filter), or a business question with its canonical SQL (kind=question). It resolves the target, computes the canonical qualifiedName, and reports action: create means a new row, update means this identity already exists and the write lands on it, revive means it exists but was archived. If the user asked to draft or review it first, call with mode=draft, show the preview, and call again with mode=execute once they approve — a tool-permission prompt is not approval. The identity is derived from the content the way Atlan's own miner derives it, so re-writing the same content converges with the mined row instead of duplicating it. Writes are attributed to you and read back to confirm: the result carries guid, qualified_name and status; filter and question add verbatim_ok (the identity text landed exactly as sent) and changed (false when the row already held exactly this), join adds edges_verified. join: source_qualified_name, joined_qualified_name, column_pairs, join_type, cardinality, when_to_use, and optionally origin, evidence, confirmed_by. filter: column_qualified_name, operator (the identity), predicate_sql, when_to_use, common_values, dataset_qualified_name. question: dataset_qualified_name, question, canonical_sql. All qualifiedNames are full Atlan qualifiedNames including the connection prefix.

create_sql_insight

Permanently delete one or more custom metadata sets and all their attribute values from assets. Irreversible. Use mode='draft' when the user asked to draft, preview, or review the change before it is applied.

delete_custom_metadata_set_tool

Delete data quality rules by their GUIDs. Use mode='draft' when the user asked to draft, preview, or review the change before it is applied.

delete_dq_rules_tool

List a type's valid attributes (with descriptions) and relationships as camelCase field names — use this to pick the exact names to pass to get_assets' `attributes` (own scalar fields) and `relationships` (related entities) when a request needs type-specific fields (e.g. SQL-Intelligence joins/filters).

describe_asset_type_tool

PRIMARY search tool for the Atlan catalog — use this for ALL structured catalog searches (by name, type, hierarchy, governance, tags, dates), in preference to search_assets_tool. Use semantic_search_tool only for pure free-text semantic discovery with no structured filter. Each asset in the result carries a `url`: when you name an asset in your answer, link the name to that `url` exactly as given.

fluent_search

Discover the fields and operators a fluent_search query can use, live from this tenant's schema.

fluent_search_schema

Fetch SVG icons for asset types. Internal tool used by widgets.

get_asset_icons

Retrieve full details for one or more assets you already have GUIDs for — typically from a prior search tool call. Does not discover assets by name or keyword — search first and read `guid` off its results. Never invent a GUID, and never pass a qualifiedName (the slash-separated path like 'default/snowflake/123/DB/SCHEMA/TABLE') where a GUID belongs. `relationships` and `attributes` names are camelCase, matching Atlan's field naming. There is no `include_readme` parameter here (that flag is search-only); to fetch an asset's README, pass 'readme' in `relationships` (or `attributes`). Returns catalog METADATA only — the file bodies of skills, knowledge files, and other object-store artifacts are never in these attributes; to read that content use read_artifact. Each asset in the result carries a `url`: when you name an asset in your answer, link the name to that `url` exactly as given.

get_assets_tool

Retrieve the certified context curated FOR THIS ESTATE before writing SQL to answer a business-metric question about its governed tables. One call returns one block: (1) the estate's analyst conventions — population filters, metric formulas, naming rules and platform vocabulary its analysts apply; (2) the top matching certified metric INDEX cards — each with the concept's decisions, ONE canonical execution-verified SQL, and [[guid]] wikilinks to its variants and sibling metrics — plus a shortlist of other matches; (3) real analyst queries from the estate's own history (CANONICAL entries are the estate's own recipes). Prefer this over general asset search when the question is about business metrics on certified tables: it returns curated definitions, not catalog matches. A tenant may enable several glossaries for this tool (one per estate); by default all are searched and every served object is labelled with its glossary — pass `slug` to narrow to one. Glossaries a customer authored in the ordinary way (no certified structure) come back as GLOSSARY DEFINITIONS — definition, synonyms, linked tables, README — which are vocabulary and business rules to interpret the question with, not verified SQL: derive the query from the schema. Then open any [[guid]] with get_metric_tool when a different variant's grain/filter fits better — or, for questions spanning several metric families (e.g. a full funnel), EACH relevant sibling index. If the result carries empty_result/guidance or truncated/truncation_note, follow it.

get_context_tool

List workspace user groups (collections of users for permissions and ownership) and, optionally, their members. Use for group and membership questions ("which groups exist?", "who is in X?"). Filter by name_filter (substring) or group_id (exact GUID); set include_members=true only when you need the user list. Members are paged — use member_offset + member_limit for large groups. For individual users use get_users.

get_groups_tool

Get a presigned PUT URL for uploading a binary file to the Atlan knowledge store. Use for large files (>5 MB) or binary formats (PDF, docx, images) when direct HTTP PUT is available. Returns upload_url (PUT the file bytes here) and upload_ticket (pass unchanged to register_knowledge_file_tool). For text or files under 5 MB, use save_knowledge_content_tool instead.

get_knowledge_upload_url_tool

Get ONE governed metric DEFINITION — the glossary term's README, not a computed value — by GUID (from a [[guid]] wikilink) or exact name. An INDEX term returns the concept's definition, population/grain decisions, ONE canonical execution-verified SQL, and 'Related variants' / 'Related metrics' wikilinks. A VARIANT term returns that variant's exact definition (a Power BI DAX measure or a real analyst query). A term from a glossary the customer authored themselves (kind 'plain') returns its definition, synonyms, linked tables and README — vocabulary, not verified SQL. Prefer this over get_assets_tool for glossary terms: it returns the README content and navigation wikilinks directly, ready to use, instead of raw asset attributes. The result names the glossary the term came from (glossary, glossary_slug); when a NAME exists in several enabled glossaries the result says so (ambiguous, candidates) and its guidance tells you how to pick another with `slug` or the candidate's id. Open a variant only when its grain, filter or time-intelligence fits the question better than the index's primary example; for questions spanning several metric families, open EACH relevant sibling index — their populations and channel scopings differ. If the result carries empty_result, follow its guidance back to get_context_tool.

get_metric_tool

Find workspace users by name, email, or username, or fetch one user by their exact GUID (user_id) — for owner/steward assignment and directory lookups ("who is X?", "find the user alice@acme.com", or resolve an owner GUID to a user). For groups use get_groups; for tags/glossaries/domains use resolve_metadata.

get_users_tool

Open the drag-and-drop upload widget for uploading files to the Atlan knowledge store. Pass folder_name to upload into a specific folder (e.g. 'SOP', 'Finance'). Defaults to the provider's default folder.

knowledge_upload_file_manager

List knowledge files (widget-internal).

knowledge_upload_list_files

Read a knowledge file by name (widget-internal).

knowledge_upload_read_file

Store uploaded files into the named knowledge folder.

knowledge_upload_store_files

List knowledge folders (no folder_name) or files in a specific folder (with folder_name). Call this before uploading to show existing contents and get user confirmation. Without folder_name: returns all folder names and file counts. With folder_name: returns files with path, guid, and optionally is_indexed + chunk_count when include_indexed_status=true. Do not call this before search_knowledge_tool — search works across all folders without needing folder names first.

list_knowledge_tool

Add or remove announcements (information, warning, issue) on assets. Use mode='draft' when the user asked to draft, preview, or review the change before it is applied.

manage_announcements_tool

Manage asset lifecycle: archive, restore, or permanently purge assets. WARNING: PURGE cannot be undone. Use mode='draft' when the user asked to draft, preview, or review the change before it is applied.

manage_asset_lifecycle_tool

Create or remove lineage between assets by connecting them through a Process. 'create' upserts the edge (re-running the same input→output edge updates in place, it does not duplicate); 'delete' permanently removes (hard-deletes) the edge this tool created — this cannot be undone. Resolve asset names to GUIDs with search first. ALWAYS leave mode unset — both create and delete are drafted in the review widget for the user to approve; never pass mode='execute' yourself.

manage_asset_lineage_tool

Run a read-only SQL query against a connection and return rows from its source database. Queries real data in the source, not Atlan metadata. Schema structure — tables, columns, data types — is already complete in the catalog (get_assets_tool / search_assets_tool); SQL is for data values, aggregates, and previews. Before querying, discover the exact connection, schema, table, and column names with semantic_search_tool or get_assets_tool — do not guess identifiers. Write SQL in the dialect of the connection's warehouse (Snowflake, Redshift, BigQuery, Databricks, …), inferred from connection_qualified_name. Requires query permission on the connection; if you get a permission or credential error, do not retry — answer from metadata (semantic_search_tool / get_assets_tool) instead. Returns up to 100 rows; add a LIMIT or narrow the query if you need fewer or different rows.

query_assets_tool

Browse and read Atlan's ARTIFACT catalog: skills (SKILL.md + supporting files), knowledge files, and agents (context repos). Artifacts are stored separately from data assets — questions like 'what skills do we have?' or 'list our knowledge files' are answered by calling this tool with NO arguments, which returns all three kinds with names, descriptions, and GUIDs. Never answer artifact-discovery questions with search_assets_tool or semantic_search_tool: the metadata catalog does not reliably surface artifacts, and an empty result there does not mean none exist. With a GUID, qualifiedName, or exact artifact name: returns the artifact(s) as markdown with file content inlined. Multi-file skills inline SKILL.md and return a manifest for the rest (fetch via artifact_path). For an artifact's catalog metadata (owners, tags, certificate) use get_assets_tool — metadata never contains file content. Content is capped per artifact; a truncated artifact says so — ask before re-fetching in full via include_content=false (the download link is for your use, not for display).

read_artifact

Read the complete text of a specific knowledge file. Prefer search_knowledge_tool for answering questions — it returns relevant passages in far fewer tokens. Use this when you need the full document: summarising all content, translating the whole file, structural analysis, or when the file is not yet indexed (is_indexed=false from list_knowledge_tool). If truncated=true in the response, offer to search specific sections with search_knowledge_tool instead.

read_knowledge_file_tool

Register a knowledge file in Atlas after a successful presigned PUT upload. Pass the upload_ticket from get_knowledge_upload_url_tool unchanged. Makes the file discoverable via list_knowledge_tool and readable via read_knowledge_file_tool. To link the registered file to the assets it describes, call update_assets_tool with knowledge_files using the returned file guid.

register_knowledge_file_tool

Remove an Atlan tag from one or more assets. Pass a dict for single asset, list for batch. Use mode='draft' when the user asked to draft, preview, or review the change before it is applied.

remove_atlan_tag_tool

Remove a custom metadata set from an asset. Use mode='draft' when the user asked to draft, preview, or review the change before it is applied.

remove_custom_metadata_tool

Remove (archive) attributes from an existing custom metadata set. Archived attributes are soft-deleted and their values cleared from all assets. Use mode='draft' when the user asked to draft, preview, or review the change before it is applied.

remove_attributes_from_cm_set_tool

Resolve Atlan tenant-level metadata definitions by name or keyword and return their exact names + GUIDs: Atlan tags (classifications) and custom metadata (business_metadata) sets. Use it to discover which of these exist, or to resolve one's identifier before another call (e.g. tagging, editing custom metadata). For glossaries, terms, categories, data domains & products, and any data asset (tables, columns, dashboards) use semantic_search or search_assets — those are persona-scoped; for people use get_users; for groups use get_groups.

resolve_metadata_tool

Save content as a knowledge file in Atlan. Routes through Heracles (single multipart POST to /api/service/knowledge/upload) so the AI gateway key is injected and Wisdom runs TurboPuffer enrichment inline - parse, chunk, embed, upsert. After this call the file is immediately searchable with search_knowledge_tool. Supports text (md/txt/json/yaml via content) and binary files up to 10 MB (pdf/docx via content_base64). Identical content is deduplicated by SHA-256 hash. To link the saved file to the assets it describes, call update_assets_tool with knowledge_files using the returned file guid.

save_knowledge_content_tool

Schedule data quality rule execution on assets. Use mode='draft' when the user asked to draft, preview, or review the change before it is applied.

schedule_dq_rules_tool

Find Atlan assets by attribute value, tag, domain, or glossary term — use this when you don't already have a GUID. If fluent_search is present in your tool list, prefer it for structured catalog searches — it is the primary search tool. Follow up with get_assets, using guid from these results, for full detail on any hit — including a table's complete column list — this tool returns a curated subset per result, not everything. Skills, knowledge files, and agents live in the artifact catalog, not here — to list or read them use read_artifact; an empty result here does not mean no artifacts exist. (To filter artifact ASSETS by owner or tag, the typeNames are Skill, SkillArtifact, KnowledgeFile, ContextRepository, ContextArtifact.) Each asset in the result carries a `url`: when you name an asset in your answer, link the name to that `url` exactly as given. NOTE: on this tenant, prefer fluent_search for catalog search — it is the primary structured-search tool here and search_assets_tool is being retired.

search_assets_tool

Search Atlan's customer-facing documentation and return an LLM-generated answer with source citations. Use for how-to questions about Atlan features — not for searching data assets (use semantic_search_tool for that).

search_atlan_docs_tool

Search knowledge files semantically to answer questions or find relevant context. Use this as the first and primary knowledge retrieval step - before listing files, before reading files. Returns focused text chunks (400-512 tokens each) with source citations (file name, folder, page number, document title). If results are insufficient, call list_knowledge_tool to see what folders exist, then offer to add missing content with save_knowledge_content_tool. To link a file you found to an asset as grounding context, call update_assets_tool with knowledge_files.

search_knowledge_tool

Search the Atlan data catalog with a natural-language question. Use fluent_search first for any question that names an asset type, name, owner, certification status, domain, tag or connector; call semantic_search_tool only when the question has no such identifier, or after fluent_search returned nothing. An internal planner reads the full question, breaks it into filters, and runs a hybrid keyword+semantic search over catalog metadata — so ask in plain language rather than keywords. It can filter and combine: asset type (table, column, view, dashboard, glossary term, domain, data product, …), connector, database/schema, domain/product, owner, certification (verified/draft/deprecated), tags/classifications, custom metadata, glossary terms, dates, and row/column counts. Supports counts and breakdowns ("how many X", "X by Y"). Searches catalog metadata, not row-level data values. To look up a specific user or group, use get_users_tool or get_groups_tool. Skills, knowledge files, and agents live in the artifact catalog, not here — to list or read them use read_artifact; an empty result here does not mean no artifacts exist. Response includes assets, total_count, has_more, and next_offset. When has_more is true, call again with the same user_query and offset=next_offset to get the next page. Each asset in the result carries a `url`: when you name an asset in your answer, link the name to that `url` exactly as given.

semantic_search_tool

Record the user's feedback (thumbs up/down + optional comment) about a prior answer or result.

submit_feedback_tool

Traverse upstream or downstream lineage from one or many assets. Pass `guids` with one or more starting assets — a single guid is fine (guids=["<uuid>"]); several roots are traversed and merged into one graph (each node carries ``reached_by`` listing which roots reach it). Returns a widget-ready graph with data assets connected by lineage edges. Process connector nodes are bridged in the graph (A -> B instead of A -> Process -> B); their entity data is exposed in the ``process_map`` field and each relation carries ``via_process_guids`` identifying which ETL/transformation processes produced the connection. The lineage widget supports progressive loading — leaf nodes show an expand button that fetches deeper lineage on demand, so keep `limit` small (5-15) for a responsive interactive view. To retrieve a whole graph in one call (impact analysis, export), set immediate_neighbors=false with a higher `depth` and `limit` (up to 100). Each asset in the result carries a `url`: when you name an asset in your answer, link the name to that `url` exactly as given.

traverse_lineage_tool

Update attributes on one or more assets. Each item identifies its asset by guid (preferred) or qualified_name plus the fields to change. To change several fields on the same asset, set them all in one update dict in a single call (one approval applies them together) — never split an asset's fields across calls. Use mode='draft' when the user asked to draft, preview, or review the change before it is applied.

update_assets_tool