Connie

Browse records in an object. Lists Records in a Connie Object, paginated via limit and offset, so you can INSPECT the actual data (Field names + sample values) before building or wiring a Workflow. Every row includes reserved Record metadata `_id`, `_created_at`, and `_updated_at`; `_updated_at` is when that Record itself was last modified, distinct from user-defined Fields such as `Last Activity` or `Last activity date`. Reads the WHOLE Object and cannot be scoped to a List — for a question about a List use search-record or the object-search-* read with listId, which is a different Record set. Cell text is truncated to maxCellChars (default 200) to keep the response context-safe; pass maxCellChars:0 for full values. Address the Object by UUID or by NAME (a non-UUID tableId is resolved automatically; a Workflow ID also works, since Workflows and Objects share one ID). Read Records in an Object (paginated, truncated by default to avoid context bloat) to see what the real data looks like. Matches "show all Records", "what's in the Object". Use search-record to filter, or get-record for one Record by UUID.

object-list-records

View a workflow’s recent runs and their results. Returns recent Workflow runs, newest first (limit up to 100): for ONE Workflow when flowId is given, or across EVERY Workflow the user can view when flowId is omitted (each run then carries workflowId and workflowName). Filter server-side with status (failed / succeeded / running) and since / until on the run's start time. Each run has its status, start/end times, run context, and — for failed runs — the real errorMessage pulled from the run log. This is the RUNTIME failure path: use it to explain why a run failed or which Action showed the red error icon on the canvas, as opposed to workflow-diagnose's pre-run validation. 'My last failed run' with no Workflow named = omit flowId, status:'failed', then runs[0]. RUN SCOPE IS EXACT: 'last/latest/most recent' means runs[0] from this result; pin that run id and drill into it with workflow-get-run-details. A shared Record id does not merge sibling runs. Keep the selected run on follow-ups unless the user explicitly changes it or asks to compare runs. List recent Workflow runs — one Workflow's, or the whole workspace's — with status, timing, and the actual failure reason for failed runs. Use for 'why did my Workflow fail / did it run / what failed today'.

workflow-get-runs

Find a record in a list and view its details. Fetch ONE Record by its `recordId` (from search-record / get-record), returning every Field value, relation links, AND the Lists the Record belongs to. The Object is derived from the Record — you do NOT pass an Object id. Get one Record by its `recordId` — all Field values, relations, and the Lists it's in. The Object is derived from the Record.

get-record

Read a conversation’s details and its messages, newest first, a page at a time. Fetches one external inbox conversation by its UUID, including status, labels, lead-type, and assignee, plus its non-deleted messages ordered NEWEST FIRST, up to `limit` per page (default 100). When the result has `hasMore: true`, pass its `nextCursor` back as `cursor` to read further back in time; `since` keeps only messages sent at or after a date or time, for questions about what arrived recently. A page without `hasMore` is the rest of the conversation. Resolve the conversation ID with inbox-list-conversations or inbox-search. Use inbox-summarize for a condensed AI summary rather than raw messages. Read one inbox conversation’s metadata and its messages, newest first, with a cursor to page back and a since filter.

inbox-get-conversation

Get a call's AI summary, next steps, key moments, metrics, topics, and sentiment. Returns the AI-generated insights for one call by UUID — summary text, next steps, key moments, key metrics, topics, and sentiment. Requires a callId (from calls-list). Does not include action items (use calls-get-action-items) or the raw transcript (use calls-get-transcript). Get a call's AI summary, next steps, key moments, metrics, topics, and sentiment. For follow-up tasks use calls-get-action-items; for the raw transcript use calls-get-transcript.

calls-get-summary

Save supplied search results or researched rows in a Connie List with their sources and review status. This writes workspace data; it does not fetch new results from a provider.

list-save-results

Find companies by name or field values. Search Connie's Companies Object by `query` (free text, best for names) or `filters` (Field → value, OR-ed across Fields; also {"$contains":"text"} and {"$is_empty":true}/{"$is_not_empty":true}). The schema lists this workspace's Companies Fields (e.g. Company (name), Website URL, Stage, Owner). Omit both to list a page (max 100). SEARCH-FIRST — HARD: run this before any update/delete. Act on a result's `_id` — never a row number or an invented id. Exact name matches come first. Before creating a company, ONE search by name (or email) is the whole duplicate check: zero results means there is no duplicate, so create it. object-create-company also refuses a same-name duplicate on its own. ZERO RESULTS: when you are looking up an existing company, stored names can hold non-breaking spaces, so retry with one word: {"Company":{"$contains":"<last word>"}}, then the first token. A duplicate check before a create is not such a lookup: its zero result is the answer, so never retry it on tokens. AMBIGUITY: a one-word match can be ambiguous: with 2+ hits, confirm with another Field (Email, LinkedIn URL, Company) or ask; never act on an unconfirmed match, and never invent companies. Ranges, dates, rankings and conditions that must all hold cannot be expressed here; do not approximate one or call the question unanswerable: use `search-record` on this Object (AND-combined `data` with $gt/$gte/$lt/$lte/$ne and $on_date/$date_gte/$date_lt, or sortBy + limit for the top N). This returns a page of Records, not totals: for counts, sums or breakdowns use `search-record` with countOnly, or the analytics tools. Search the native companies Object by name, Field value, or blank-field filter (Fields enumerated for this workspace) — NOT HubSpot/Salesforce. Run before update/delete, and once before a create as the duplicate check.

object-search-companies

Keyword-search external inbox messages by subject/body content. Full-text searches the workspace's connected external mailboxes and channels, matching the query against message subject and body and returning the matching conversations. Pass query (required) and optional limit (default 20). For browsing recent conversations without a keyword use inbox-list-conversations; to read one matched thread use inbox-get-conversation. ROUTING: find/show/manage existing threads → inbox-* tools; send/forward/schedule a new message → email-* tools. BY NAME (delete, pin, archive or label a person's conversation): results come best name match first — the whole name or a whole word, then a word that starts with the query, then a name that only contains it. If the result carries nameMatch.ambiguous, ask the person which of those contacts they mean before changing anything; otherwise act on the FIRST result. DELETE EMAIL BY NAME: inbox-search {query:"<person>"} → inbox-delete {conversationId} ONCE; its approval card names the conversation. Zero matches → "No emails found from <person>." Use inbox-delete (writes the inbox DB directly), not email-* atomics. "delete ALL emails from X" → inbox-bulk-action once. Keyword-search external inbox messages by subject/body content — 'find emails about X', 'messages from Acme'. The external inbox, NOT this assistant's chat history.

inbox-search

Find people by name or field values. Search Connie's People Object by `query` (free text, best for names) or `filters` (Field → value, OR-ed across Fields; also {"$contains":"text"} and {"$is_empty":true}/{"$is_not_empty":true}). The schema lists this workspace's People Fields (e.g. Person (full name), First name, Last name, Email, Phone, Job title, Company, LinkedIn URL, Stage). Omit both to list a page (max 100). SEARCH-FIRST — HARD: run this before any update/delete. Act on a result's `_id` — never a row number or an invented id. Exact name matches come first. Before creating a person, ONE search by name (or email) is the whole duplicate check: zero results means there is no duplicate, so create it. object-create-person also refuses a same-name duplicate on its own. ZERO RESULTS: when you are looking up an existing person, stored names can hold non-breaking spaces, so retry with one word: {"Person":{"$contains":"<last word>"}}, then the first token. A duplicate check before a create is not such a lookup: its zero result is the answer, so never retry it on tokens. AMBIGUITY: a one-word match can be ambiguous: with 2+ hits, confirm with another Field (Email, LinkedIn URL, Company) or ask; never act on an unconfirmed match, and never invent people. Ranges, dates, rankings and conditions that must all hold cannot be expressed here; do not approximate one or call the question unanswerable: use `search-record` on this Object (AND-combined `data` with $gt/$gte/$lt/$lte/$ne and $on_date/$date_gte/$date_lt, or sortBy + limit for the top N). This returns a page of Records, not totals: for counts, sums or breakdowns use `search-record` with countOnly, or the analytics tools. Search the native people Object by name, Field value, or blank-field filter (Fields enumerated for this workspace) — NOT HubSpot/Salesforce. Run before update/delete, and once before a create as the duplicate check.

object-search-people

Search only the Connie actions individually listed on this ChatGPT connector. Returns matching action identifiers and descriptions; this does not enable any other operation.

search-tools

Run one action by exact identifier, only if it is also individually listed on this ChatGPT connector. The action keeps its own schema, permissions, side effects, and billing; unlisted identifiers are refused.

use-tool