cancel_workflow_run
ChatGPTIdempotently cancel an owned active run. Existing partial outputs remain available.
Overview
Convoke gives your agents access to an up-to-date biopharma intelligence database, a structured knowledge graph spanning drug development programs and their catalysts. Catalysts include upcoming clinical trial milestones and regulatory or commercial events. Search programs by drug, target or disease name and resolve to canonical identifiers.
Tools
create_grid
ChatGPTCreate a grid in an authorized writable workspace. Returns its durable ID.
create_grid_columns
ChatGPTCreate typed columns with configuration and per-item receipts.
create_grid_rows
ChatGPTCreate up to 100 rows with hierarchy and per-item receipts; inspect failures before retrying.
create_monitor
ChatGPTCreate an active monitor from an overview, criteria, and optional priority tiers. Write an overview of what to track and why, criteria defining which findings to include or exclude, and optional priority tiers (1 highest). Preserve the user's priorities and exclusions; do not invent restrictions. Keep resolved entity_ids aligned with the written scope. Criteria filter findings; entity_ids drive retrieval. Pass resolved entity IDs matching the written scope. get_drug_programs returns drug_id, indication_id, and entity_resolution for supplied name filters, including organizations and targets. Omit enabled_dimensions to infer sources from the brief. Call when asked to create, save, or activate; present drafts in conversation without saving. Uses the default schedule. Returns the saved monitor ID, URL, and configuration; findings arrive asynchronously. Each call creates a new monitor. After uncertain completion, discover/read before retrying.
create_report
ChatGPTSave supplied HTML as a report in a writable workspace. This performs no research or generation.
delete_grid
ChatGPTPermanently delete a grid with all its rows, columns, and cells. Cannot be undone. Call only when the user asks to delete this grid by name or ID. Requires editor access. Returns grid_id, name, and deleted.
delete_grid_columns
ChatGPTPermanently delete 1-100 columns and every cell in them. Cannot be undone. Call only when the user asks to delete these columns; if which columns is unclear, confirm first. A column that a remaining computed column uses is refused; delete both together. Unknown IDs reject the whole call. Requires editor access. Returns grid_id, column_ids, and deleted_count.
delete_grid_rows
ChatGPTPermanently delete 1-1,000 rows, their descendant rows, and every cell in them. Cannot be undone. Call only when the user asks to delete these rows; if which rows is unclear, confirm first. Unknown IDs reject the whole call. Requires editor access. Returns grid_id, row_ids, and deleted_count, which includes descendants.
delete_monitor
ChatGPTDelete an org-owned monitor by ID. Soft-deletes the monitor and initiates schedule and external-monitor cleanup; running work may finish asynchronously. Call only when the user requests deletion.
get_document_download
ChatGPTGet a one-hour attachment download link to the original Vault file in an accessible workspace. Returns filename, download_url, expires_at (UTC), media_type and size_bytes. Request again after expiry.
get_drug_programs
ChatGPTFilter drug programs. Select fields and follow next_offset; inspect complete and entity_resolution.
get_events
ChatGPTQuery Catalyst Calendar events with source attribution and explicit page coverage.
get_grid_cells
ChatGPTRead selected cells. Choose fields for values, explanations, sources, reviews, and provenance. Provenance includes who last edited each value. To hyperlink a cell, use https://platform.convoke.bio/grids/{grid_id}?row={row_id}&column={column_id}. Use the returned grid, row, and column IDs exactly.
get_grid_download
ChatGPTGet a one-hour whole-grid download as tidy CSV (default) or structured JSON. CSV columns: row_id,row_name,column_id,column_name,value, including unset cells. JSON includes grid and column prompts, row hierarchy, and selectable cell fields (same options as get_grid_cells); fields applies only to JSON. In JSON, join each cell's row_id and column_id to rows[].id and columns[].id (names may repeat), follow parent_row_id for subrows, and order by position. Returns filename, download_url, expires_at (UTC), media_type and size_bytes. 100 MiB / 120 seconds; oversized/failed exports return no partial download. Request again after expiry.
get_grid_enrichment_status
ChatGPTInspect current enrichment state for a grid.
get_grid_structure
ChatGPTRead visible grid rows, columns, hierarchy, and configuration to select IDs for cell reads. For a large or unfamiliar grid, start with mode=compact: grid identity and row, root row, maximum depth, column, in-progress and failed cell counts, without listing rows or columns. Then page rows and columns with search_grid_rows and search_grid_columns (pattern "."), or read mode=full.
get_monitor
ChatGPTRead an org-owned monitor's configuration, scope, and schedule.
get_monitor_findings
ChatGPTRead findings with selected fields and page coverage. Exact finding_ids ignores limit/offset (max 200).
get_project_file_download
ChatGPTGet a one-hour attachment link to a file in a project you own. Supply project_id and path. path is project-relative or /project/ absolute. Hidden state is excluded except saved tool outputs. Returns filename, download_url, expires_at (UTC), media_type and size_bytes. Request again after expiry.
get_report_download
ChatGPTGet a one-hour attachment download of the complete report as Markdown, with no text clipping. Returns filename, download_url, expires_at (UTC), media_type and size_bytes. 100 MiB / 120 seconds; failed exports return no partial download. Request again after expiry.
get_saved_views
ChatGPTDiscover accessible saved views for a required tracker_type. Omit saved_view_id to list summaries; provide it to read one view with its view_config. The requested tracker must be enabled. Results are budgeted to approximately 28,000 characters. When truncated=true, continue with offset=next_offset. Limit defaults to 50 (maximum 100).
get_transcripts
ChatGPTRead exact transcripts. Select text_offset/text_limit for text ranges and follow each next_text_offset. Ranged requests return fallback_text; full requests retain speaker paragraphs. Retry only failed IDs.
get_workflow_run
ChatGPTInspect an owned run. Follow next_metadata_offset; concatenate metadata_json pages before parsing.
list_monitor_runs
ChatGPTRead monitor execution history; follow next_offset for more runs.
list_monitors
ChatGPTList the org's monitors, newest activity first. Output per item: id, artifact_type, name, plus alert_data (status, last_run_at, schedule_crons). Use get_alert to read the monitor's scope. Envelope: total_count, returned_count, truncated, next_offset, error (null on success).
list_tags
ChatGPTDiscover tag usage aggregated across accessible workspaces, or narrowed by workspace_id. Use the same workspace_id with search_artifacts to resolve a tag in that workspace.
list_workflow_runs
ChatGPTList recent owned workflow runs.
list_workspaces
ChatGPTList accessible Document Vault workspaces, roles, and writable destinations.
move_grid_rows
ChatGPTMove 1-10,000 sibling rows, including their subtrees, within the same grid. Use to move a few rows or change their parent; to sort or fully rearrange siblings, use reorder_grid_rows. row_ids sets their destination order. placement=before/after puts them next to target_row_id, under its parent. placement=start/end puts them first or last among parent_row_id's children (null: root rows). Preserves cells and requires editor access. Missing IDs, mixed source parents, invalid anchors, or cycles reject the whole call. Read structure and reconcile before retrying after uncertain completion. Returns grid_id, row_ids, parent_row_id, and changed_count (including shifted siblings); a no-op returns 0.
query_documents
ChatGPTSearch visible document text; control ranked chunk count with max_results (1-100). total_count counts retrieved candidates, not all corpus matches. Increase max_results to broaden retrieval. Use read_document for surrounding source text and full tables. retrieval_context is indexing context, not necessarily verbatim source text.
read_document
ChatGPTRead extracted source Markdown for a document_id from search_artifacts or query_documents. Follow next_text_offset for more text; pass content_sha256 as expected_content_sha256 on later pages to detect edits. Concatenate adjacent slices if a table spans the boundary.
read_report
ChatGPTRead text ranges (default 200 lines, max 500), outline pages, or table rows (offset/limit, max 200). Discover table_ref in outline mode. Follow next_start_line for text or next_offset for outlines/tables.
rename_grid
ChatGPTRename a grid. Requires editor access. Returns grid_id and name.
reorder_grid_columns
ChatGPTPersist the desired order of every column in a grid (1-1,000 unique column IDs). Read get_grid_structure first. Preserves column configuration and cells. Requires editor access. Missing or extra IDs reject the whole call; reload the structure and retry. Returns grid_id, ordered_count, and changed_count; repeating the current order changes nothing.
reorder_grid_rows
ChatGPTPersist the desired order of every sibling under one parent (1-10,000 unique row IDs). Use to sort or fully rearrange one parent's children. To move a few rows or change their parent, use move_grid_rows. Read get_grid_structure first. parent_row_id=null selects root rows; otherwise supply every direct child of that parent. Preserves parents, subtrees, and cells. Requires editor access. Missing, extra, or wrong-parent IDs reject the whole call; reload the structure and retry. Returns grid_id, ordered_count, and changed_count; repeating the current order changes nothing.
search_artifacts
ChatGPTSearch visible artifact names; optionally narrow by workspace_id. Results include workspace_id. Select fields and use limit/offset and next_offset to page.
search_congress_presentations
ChatGPTSearch presentations; page with next_offset. Request private_document_ids to query accessible attachments.
search_disclosures
ChatGPTSearch ranked transcript and SEC passages; max_results controls the returned size.
search_grid_columns
ChatGPTMatch PostgreSQL regex against column names/instructions, not cell values. Case-insensitive by default. Pattern: 1-500 characters. Page in stable ID order using limit (1-200) and next_offset. Pages reflect current data; complete means this result contains all matches.
search_grid_rows
ChatGPTMatch PostgreSQL regex against row names/descriptions, not cell values. Case-insensitive by default. Pattern: 1-500 characters. Page in stable ID order using limit (1-200) and next_offset. Pages reflect current data; complete means this result contains all matches.
start_grid_enrichment
ChatGPTStart selected grid enrichment and return a receipt. Inspect status before starting more work.
stop_grid_enrichment
ChatGPTStop active grid enrichment. Partial completed results remain available.
update_grid_cells
ChatGPTAppend complete cell versions with explicit value and evidence choices. Supply value or copy_previous_value=true. Supply both explanation and sources, or copy_previous_metadata=true. Copy flags cannot accompany replacement fields for that component and require a matching non-null expected_value_id. Supply at least one replacement component. Null value/explanation and [] sources clear them. Supply evidence supporting changed values; copy only evidence that still applies. A stale version rejects the whole batch; reload before editing. History is retained.
update_grid_columns
ChatGPTRename columns, rewrite their prompts (instructions), change their output type, or retarget them. Only the fields given change; get_grid_structure reads the current values. A new prompt applies to future enrichment runs, and a type change recasts existing values. Computed columns are edited in the app.
update_grid_instructions
ChatGPTReplace a grid's instructions, the guidance every cell agent gets on future enrichment runs. get_grid_structure reads the current instructions. Null or blank text clears them.
update_grid_rows
ChatGPTRename 1-100 rows or rewrite their descriptions, in one transaction. Only the fields given change; get_grid_structure or search_grid_rows reads the current values. Names and descriptions guide future enrichment runs; existing cells are not re-researched. Unknown IDs reject the whole call. Requires editor access. Returns grid_id, updated_count, and each row's id, name, and description.
update_monitor
ChatGPTUpdate an org-owned monitor by ID. Read current configuration with get_monitor. Omitted or null fields stay unchanged. Supplied criteria, prioritization, and entity_ids replace their lists. Empty criteria or prioritization clears them, but scope must stay nonempty. Brief edits preserve entity scope unless entity_ids is supplied. Set status to paused to stop updates or active to resume; resuming initiates the existing backfill. Repeating the saved status does not start another backfill. Legacy monitors only support name, scope, and status edits. The receipt confirms saved changes; running work and follow-up synchronization may finish asynchronously. Call only when the user requests changes.
query_catalyst_calendar
Claudequery_program_tracker
ClaudeCapabilities
Writes
Example Prompts
Click any prompt to copy it.
