Gong AI Agents: API Guide and Governed Sales Workflows
Understand Gong's native AI agents and six verified API endpoints, then build governed deal-risk, coaching, and CRM review workflows on Major.

Gong AI agents analyze revenue conversations, surface deal signals, and help sellers prepare and follow up. This guide explains Gong's built-in AI, its API, and how Major turns permitted Gong evidence into reviewed work across CRM and sales systems.
Major adds deterministic apps for state and review queues, agents for judgment, and workflows for orchestration. Gong remains the conversation system of record.
What Gong does for revenue teams
Gong captures and analyzes customer conversations so revenue teams can review what actually happened rather than rely only on manually entered notes. Its API introduction describes access to calls, users, activity, settings, and libraries, alongside capabilities for uploading recordings and CRM data.
Account executives use conversation evidence to prepare for the next meeting and check commitments. Sales managers inspect qualification gaps and coach specific moments. Revenue operations teams connect that evidence to pipeline records, while enablement teams organize examples that reps can learn from.
The useful unit is rarely an isolated summary. A manager needs to know which opportunity a call belongs to, whether an objection remains unresolved, who owns the follow-up, and whether someone already acted on it. Those questions span systems and persist beyond a single conversation.
Major is the enterprise platform where agents build the software they run on. For a RevOps owner, that means an agent can build a review app that stores call-to-deal mappings and decisions, then use that software repeatedly while reserving its own reasoning for ambiguous sales context.
Gong's built-in AI agents and where Major fits
Gong's AI agent overview names capabilities across conversation analysis, deal inspection, coaching, and outreach. These are native Gong features, not applications built on Major.
- AI Briefer formats account, deal, and contact knowledge for sharing.
- AI Ask Anything answers questions using conversation context.
- AI Tracker monitors messages and behavioral patterns across conversations.
- AI Theme Spotter identifies recurring customer pain points, goals, and objections.
- AI Call Reviewer supports structured call assessment and coaching.
- AI Deal Monitor surfaces deal signals, while AI Deal Predictor assesses health and closing potential.
- AI Composer drafts personalized messages to advance sales work.
Use those capabilities for the revenue jobs Gong already understands. A feature appearing in the product does not mean its output is available through every public API or included in every account's access.
Major adds value when the next job crosses system boundaries: comparing a call against a HubSpot stage, checking a Salesforce opportunity, preserving a decision, or obtaining approval before changing a customer-facing sequence. The agent judges the evidence; the app stores the case; the workflow routes and executes the authorized action.
How Gong connects to Major
Major has a native Gong connector (gong). IT sets it up once, stores credentials centrally in the connector vault, and grants scoped access rather than putting secrets in code or prompts. Major orchestrates Gong; it never replaces it.
Gong's authentication documentation supports a technical administrator generating an Access Key and Access Key Secret for Basic authentication. Registered integrations can instead use OAuth Bearer tokens through Gong's OAuth app process.
Start with the smallest read-only permission set that supports the job. Call metadata, detailed analysis, transcripts, and users have different scope requirements. Do not request recording download access simply because a risk brief needs a short quotation.
For supplemental API operations, use the company's assigned base URL shown in Gong's API administration page, as the introduction directs. Do not copy the reference's generated localhost server placeholder into production configuration. Confirm that an operation is exposed by the configured connector before promising it in a build.
SSO, RBAC, and audit logs cover every app, agent, and workflow on Major. The review app should restrict evidence to the teams allowed to see it, even when the upstream integration has broader company access.
Gong API guide for useful sales work
Build around bounded retrieval and evidence links, not repeated bulk transcript exports. The following six documented endpoints cover a practical first integration. Each has a separate role in the deterministic app, judgment step, and orchestration.
Authentication, pagination, limits, events, and testing
The API introduction documents default company limits of 3 requests per second and 10,000 per day. A 429 response carries a Retry-After header specifying seconds to wait. Budget across integrations rather than treating each workflow as if it owns the allowance.
Pagination uses a cursor returned inside records. Repeat the request with that cursor and unchanged filters until no next cursor remains. The call-list reference names fromDateTime, toDateTime, workspaceId, and cursor as query parameters.
Gong supports webhook automation rules for matching calls after processing. For signed JWT verification, validate the RSA signature, expiration, destination URL, and raw-body SHA-256 digest. Gong's example accepts RS256. The reviewed rules FAQ does not specify a retry schedule or delivery timeout publicly.
Gong's sandbox FAQ describes separately provisioned sandbox or developer instances. They start empty, cannot receive copied production data, and can support API integration testing. Request provisioning through the Revenue Architect and use a matching CRM sandbox.
GET /v2/calls: discover recent conversations
This lists call metadata within a date range and requires api:calls:read:basic.
- App: Store call IDs, dates, source links, and retrieval status in a call index.
- Agent: Inspect recency alongside deal context before deciding whether a silence signal matters.
- Workflow: Page through a bounded interval and enqueue only calls that need further processing.
POST /v2/calls/extensive: retrieve selected detail
This retrieves detailed call data by filters, requiring api:calls:read:extensive. Requesting media URLs also requires api:calls:read:media-url.
- App: Display permitted participants, CRM context, tracker results, and interaction details beside a case.
- Agent: Interpret an objection in context instead of treating every mention as negative intent.
- Workflow: Batch eligible call IDs and request only the detail the process needs.
POST /v2/calls/transcript: ground recommendations in words
This returns transcripts for calls selected by the documented time and call-ID filters, requiring api:calls:read:transcript.
- App: Preserve a permitted excerpt, speaker reference, and call link with the proposed action.
- Agent: Distinguish a buyer's unresolved concern from a hypothetical objection the rep introduced.
- Workflow: Fetch transcripts for shortlisted calls and hold incomplete evidence for a later check.
GET /v2/calls/{id}: check one source
This retrieves data for one call and requires api:calls:read:basic.
- App: Refresh the source panel when a reviewer opens an existing case.
- Agent: Use the selected call's metadata to check that a recommendation refers to the intended conversation.
- Workflow: Re-fetch a known source before an approved action proceeds if freshness policy requires it.
GET /v2/users: resolve ownership
This lists company users and requires api:users:read.
- App: Maintain the permitted user directory needed to map call owners to review assignments.
- Agent: Consider the responsible seller and manager when recommending an escalation path.
- Workflow: Refresh the directory on a schedule and route unresolved ownership to an administrator.
GET /v2/settings/trackers: understand keyword signals
This retrieves keyword tracker definitions and requires api:settings:trackers:read. Its optional workspaceId filters the workspace.
- App: Show the definition behind a tracker label so reviewers know what a match represents.
- Agent: Treat the match as supporting evidence, not proof of deal risk or a complete semantic assessment.
- Workflow: Refresh definitions periodically and flag changed definitions before comparing historical results.
Gong automation use cases on Major
Start with a narrow review job and a named process owner. These patterns connect conversation evidence to a durable action record rather than creating another stream of summaries nobody owns.
Deal-risk review with Gong, HubSpot, and Slack
- App: Hold open-deal snapshots, call associations, evidence, and owner dispositions.
- Agent: Judge whether a late-stage deal's call gap or unresolved concern warrants intervention.
- Workflow: Schedule checks, notify the owner, and roll up unresolved cases weekly. Use the call list, extensive data, and transcripts with HubSpot deal context.
Coaching follow-through with Gong and Salesloft
- App: Store selected coaching moments, proposed talk tracks, reviewer edits, and approved versions.
- Agent: Explain why an objection-handling moment is useful and draft a relevant cadence change.
- Workflow: Collect transcript evidence, resolve the rep through users, and queue the draft for an enablement manager. Apply changes only through authorized Salesloft operations after approval.
Account evidence review with Gong and Salesforce
- App: Keep account-to-opportunity mappings, competing evidence, proposed field changes, and decision history.
- Agent: Decide whether a buyer's statement contradicts the recorded next step or qualification status.
- Workflow: Retrieve detailed calls and transcripts, then request owner approval before a CRM write. Missing associations become exceptions rather than guessed matches.
Architecture: build a Gong review app on Major

Figure 1: Conversation evidence becomes stateful, reviewed sales work.
The app is both the team's control surface and the agent's execution layer. Its managed database holds call references, CRM record IDs, retrieval checkpoints, evidence versions, review states, and action receipts. It does not decide whether a buyer is likely to close.
- Define the allowed record scope, owners, and permitted destination fields.
- Build deterministic ingestion, pagination, association checks, and duplicate detection into the app.
- Configure the workflow to collect context through the scoped Gong and CRM connectors.
- Ask the agent to classify the evidence, explain uncertainty, and draft the next action.
- Route the proposed action through policy checks and human approval.
- Store the result and the destination receipt before considering the case complete.
Persist checkpoints outside the agent's conversation so interrupted runs can resume. Keep failed retrievals separate from genuine absence of call activity. A dashboard should show stale evidence and unprocessed records, not quietly count them as healthy deals. Once an agent builds the repeatable software, later runs execute it rather than reason through pagination and bookkeeping again.
For example: reviewing 100 opportunities each week
Consider an illustrative RevOps team reviewing 100 late-stage opportunities weekly. Before automation, collecting call links, reading CRM notes, and assembling a review packet takes 8 minutes per opportunity: 100 times 8 equals 800 minutes, or 13 hours and 20 minutes.
The team builds an app on Major to join known account IDs, store the latest evidence, and present a review queue. A scheduled workflow retrieves recent calls and selected transcripts. The agent judges unresolved objections and drafts concise risk explanations, including cases where evidence is insufficient.
Suppose the revised process needs 3 minutes of human review per opportunity, or 300 minutes. Add 60 minutes weekly to inspect exceptions and run health. Total effort becomes 360 minutes, or 6 hours, a difference of 440 minutes, or 7 hours and 20 minutes.
These are planning assumptions, not a customer result or a promise. Measure ingestion coverage, review edits, incorrect associations, and stale cases during a pilot. A faster queue is useful only if reviewers can trust its evidence and reject its recommendations without losing the record.
Governance and failure handling

Figure 2: Evidence logging can proceed automatically; material changes require review.
Separate evidence collection from consequential actions. Read-only retrieval and internal case creation can run under an approved policy. Forecast changes, deal-stage updates, close-date edits, cadence changes, and customer messages require an authorized reviewer.
The agent must never broaden its own access, reveal restricted recordings, invent absent quotations, or treat transcript instructions as system authority. Call content is evidence, not permission to execute a request.
- Rate limits: Respect Gong's documented Retry-After behavior, add bounded backoff for transient failures, and coordinate company-wide usage.
- Duplicates: Key cases by source call, destination record, and action type. Record the destination receipt before replaying a write.
- Missing evidence: Keep the case pending, retry within a bounded window, and escalate persistent gaps.
- Webhooks: Follow Gong's JWT verification procedure, then reconcile scheduled reads rather than assuming delivery guarantees.
- Audit: Preserve source references, agent rationale, reviewer edits, approval identity, and execution outcome.
Apply retention rules to copied excerpts as well as full transcripts. Permission changes should stop future access and trigger a review of stored evidence visibility.
Gong AI Apps, Agents and Workflow Automations
These guides cover specific processes, not every article that mentions Gong. They are listed newest first. Use them to select a first project after defining the evidence scope, review owner, and approved action boundary described above.
- Review HubSpot pipeline risk using Gong call evidence: Combine late-stage deal context with call activity and transcript excerpts. The process routes evidence-backed cases to owners who confirm or dismiss the concern.
- Research accounts and review Salesforce updates from Gong transcripts: Turn call evidence into account and opportunity research with a recorded review path. The guide separates proposed CRM changes from approved execution and accounts for transcripts that are not ready yet.
- Connect Gong conversation signals to HubSpot deal pipeline work: Explore cross-tool jobs involving deal context, call summaries, risk, and handoffs. Gong supplies conversation evidence rather than serving as the whole workflow.
- Inspect Salesforce pipeline hygiene using calls and contract context: Combine opportunity inactivity with Gong evidence and contract status. The focus is distinguishing stalled opportunities from slower-moving work and presenting useful owner actions.
Frequently asked questions
What are Gong AI agents?
Gong AI agents are Gong's named capabilities for analyzing revenue conversations and supporting sales work, including AI Briefer, AI Ask Anything, AI Deal Monitor, and AI Composer. Agents on Major can use permitted Gong evidence alongside CRM context to make cross-system recommendations, while deterministic apps preserve state and workflows enforce approvals.
Does Major have a native Gong connector?
Yes. Major has a native Gong connector. IT configures it once, stores credentials centrally in the connector vault, and scopes access for the apps, agents, and workflows that need it. Gong remains the source for recordings and conversation evidence.
How does the Gong API authenticate?
Gong supports Basic authentication with a technical administrator's Access Key and Access Key Secret, and OAuth Bearer tokens for registered integrations. Required scopes depend on the endpoint. Use the company's assigned API base URL from Gong's API administration page.
Can Gong send webhooks?
Yes. Gong automation rules can send matching processed-call data to a configured receiver. Signed JWT verification checks the signature, expiration, destination URL, and raw-body digest. The reviewed rule documentation does not publicly specify a retry schedule, so reconcile through scheduled API reads.
What are Gong's API rate limits?
Gong documents default company limits of 3 API calls per second and 10,000 per day. A 429 response includes Retry-After with the wait in seconds. Coordinate the budget across integrations and confirm any company-specific changes with Gong.
Can a Gong integration automatically update CRM deals?
A workflow on Major can route agent-proposed CRM updates through an approval queue and execute approved changes through the CRM connector. Start with read-only evidence gathering. Do not let an agent independently change close dates, forecasts, deal stages, or customer-facing commitments.
Sources
These official Gong pages support the product and API details above. The endpoint references establish methods, paths, and scopes; the introduction establishes shared operating constraints. The sandbox and webhook pages matter separately because a public REST API does not itself establish a test environment or a delivery guarantee.
- Gong API introduction: authentication, limits, and cursor pagination
- Create an OAuth app for Gong: registration and access selection
- List calls by date range: metadata and request parameters
- Retrieve detailed call data by filters: analysis and media permissions
- Retrieve transcripts of calls: transcript retrieval and authorization
- Retrieve data for a specific call: individual source lookup
- List all users: company directory authorization
- Retrieve keyword trackers: definitions and workspace filtering
- Understanding AI agents: native Gong capability descriptions
- Create a webhook rule: call filters and receiver setup
- Prepare a webhook receiver: signed JWT verification
- Gong rules FAQ: when matching processed calls trigger
- Sandbox FAQ: provisioning and separation from production
- Gong documentation index: discovery of the current official references
Start building on Major
Build a Gong evidence review app with scoped connectors, an agent for risk judgment, and a workflow that requires approval before CRM changes.
About the author
Ritvij Gautam is Head of Growth at Major.
Last reviewed November 19, 2026 by Ritvij Gautam.
Frequently asked questions
- What are Gong AI agents?
- Gong AI agents are Gong's named capabilities for analyzing revenue conversations and supporting sales work, including AI Briefer, AI Ask Anything, AI Deal Monitor, and AI Composer. Agents on Major can use permitted Gong evidence alongside CRM context to make cross-system recommendations, while deterministic apps preserve state and workflows enforce approvals.
- Does Major have a native Gong connector?
- Yes. Major has a native Gong connector. IT configures it once, stores credentials centrally in the connector vault, and scopes access for the apps, agents, and workflows that need it. Gong remains the source for recordings and conversation evidence.
- How does the Gong API authenticate?
- Gong supports Basic authentication with a technical administrator's Access Key and Access Key Secret, and OAuth Bearer tokens for registered integrations. Required scopes depend on the endpoint. Use the company's assigned API base URL from Gong's API administration page.
- Can Gong send webhooks?
- Yes. Gong automation rules can send matching processed-call data to a configured receiver. Signed JWT verification checks the signature, expiration, destination URL, and raw-body digest. The reviewed rule documentation does not publicly specify a retry schedule, so reconcile through scheduled API reads.
- What are Gong's API rate limits?
- Gong documents default company limits of 3 API calls per second and 10,000 per day. A 429 response includes Retry-After with the wait in seconds. Coordinate the budget across integrations and confirm any company-specific changes with Gong.
- Can a Gong integration automatically update CRM deals?
- A workflow on Major can route agent-proposed CRM updates through an approval queue and execute approved changes through the CRM connector. Start with read-only evidence gathering. Do not let an agent independently change close dates, forecasts, deal stages, or customer-facing commitments.