Apollo.io

Breadth across classes with authenticated write-back, not a single deep connector. Native Salesforce and HubSpot CRM connectivity with bidirectional sync, an integrations marketplace and partner ecosystem, mailbox connectivity to Gmail and Microsoft for sending, a built-in dialer, and a Chrome extension that carries prospecting into other web surfaces.

Apollo pulls users from Salesforce to provision its own seats, which is an authenticated read against the customer's CRM rather than a data feed. Outward, the API and MCP server create and update accounts, deals and tasks and move contacts in and out of sequences, so records are written into the customer's systems rather than held. That spans CRM, mailbox, telephony, browser and general integration classes.

Sourcedocs.apollo.io, knowledge.apollo.io and apollo.io/product/sales-engagementread 2026-09-04

Workflows are a first-class automation tier rather than a scheduler bolted onto sequences: they add contacts to sequences, assign tasks and coordinate prospecting without a rep clicking each step, with Agentic GTM and AI Sales listed among the platform modules.

The AI Assistant layers named agents over the same objects, reachable from People, Companies, Sequences, Lists, Workflows and Analytics, so a chain can run from finding target accounts through enrichment into sequence entry. The MCP server drives the same chain from outside, finding sequences and adding or stopping contacts in them.

This is a substrate the buyer assembles, not a named digital worker that owns a job end to end, which is why Apollo reads as a platform rather than an SDR.

Sourceapollo.io/product/sales-engagement, knowledge.apollo.io and github.com/apolloio/apollo-mcp-pluginread 2026-09-04

Grounded on two bodies of material, one of them the vendor's rather than the customer's. Apollo's AI research and prospecting assistance draw on its own contact and company database and on public buying signals: a large maintained corpus, but Apollo's asset, not the buyer's knowledge.

The customer's side is real but narrower: CRM fields synced bidirectionally, sequence and email history, call recordings with transcription, and account and contact records the team owns inside the platform, all of which the assistant reads when drafting. The customer cannot bring their own material to ground on. No document ingestion, retrieval index, knowledge base connection or upload of positioning, brand guidelines or product content is documented anywhere on the help center or product pages. So what the agent knows about the buyer's own business comes only from records already in Apollo.

Sourceknowledge.apollo.io data overview and apollo.io/product/sales-engagementread 2026-09-04

The default shape puts a person in the loop, and nothing gates the parts that run without one. Task management places outreach actions in a rep's queue to be worked rather than executed unattended, and AI drafts messages for a rep to review and send, so the common motion is assistive by design. Against that, sequences and workflows send on schedule once configured.

No approval step, review queue, hold, confidence threshold or content policy check is documented before an automated send or a workflow action. Permission profiles limit what each user may configure and send; that is access control, covered under security, rather than an agent guardrail.

Sourceapollo.io/product/sales-engagement and knowledge.apollo.ioread 2026-09-04

Both certifications and access controls are published, with the access surface documented in the help center. Apollo supports SP and IdP-initiated single sign-on through SAML with SCIM provisioning, configured at Settings, Security, Single sign on, with per-provider setup guides; once enabled, all other login methods are blocked. Admins can make multi-factor authentication mandatory, force logout across devices and deactivate accounts.

Permission profiles control what each user can access, configure, edit and manage, shipping with Admin, Non-admin and a Billing and seat manager profile, with feature-level permissions underneath. Apollo holds SOC 2 Type 2, covering Security, Availability and Confidentiality, and ISO 27001, both audited by A-LIGN. It also covers GDPR as processor and controller, CCPA and CPRA, and the EU, UK and Swiss Data Privacy Frameworks. A SafeBase trust center at trust.apollo.io carries a document request path. SSO and SCIM sit on the top tier.

Sourceknowledge.apollo.io security and permission profile articles, apollo.io/product/security and trust.apollo.ioread 2026-10-01

A detailed record of what happened, with nothing on why. Apollo Cockpit reporting, admin dashboards and customizable reports cover activity and outcomes. Sequence views hold a per-contact history of which step was sent when and what came back, and the dialer records and transcribes calls, so a manager can reconstruct what was done to a given prospect. None of that reaches agent behavior.

No decision trace, reasoning log or record of why a workflow selected a contact, why the assistant drafted what it drafted, or which rule fired is documented. No configuration audit trail, export or retention control appears on the help center or product pages. Customizable reports sit at the Organization tier. The internal logging and access monitoring described on the trust center are Apollo's own controls, not a trail the customer sees.

Sourceapollo.io/product/sales-engagement and knowledge.apollo.ioread 2026-09-04

Durable state is held at the record, not at the agent. Sequence state persists per contact (which step a person is on, what was sent, whether they replied, whether they should be stopped), and the MCP server works on that same state from outside.

Contact, account and task records sync bidirectionally to the CRM, and a Data Health Center tracks enrichment coverage over time. Those are business records and platform state, not agent memory.

No per-agent memory, cross-session recall between assistant conversations, learned preference carried between runs, or retention and forgetting control is documented.

Sourceapollo.io/product/sales-engagement and knowledge.apollo.ioread 2026-10-01

Multi-tenant cloud with no customer control over location or topology. No self-hosted option, private cloud, VPC, single-tenant tier or region selection is published. What Apollo publishes on data movement is legal cover rather than location control: GDPR as both processor and controller, and certification under the EU, UK and Swiss Data Privacy Frameworks.

Those frameworks govern transferring European personal data; they are not an option to keep it in region. The trust center's documents, including the ISO 27001 certificate and SOC 2 report, are available only on request, so any residency terms inside them are not public.

Sourcetrust.apollo.io, apollo.io/product/security and apollo.io/company/privacy-centerread 2026-09-04

Two kinds of packaged asset, and both are browsed rather than authored. Apollo ships prebuilt sequence templates covering emails, calls and automated tasks, with an industry template library published as a standing resource. The AI Assistant also presents named agents the user picks from a list, reached from People, Companies, Sequences, Lists, Workflows or Analytics.

A worked example shows choosing an agent such as Find target accounts, with a View all control to see the rest. Buyers can browse a catalog and select from it on both surfaces. One limit: the agent catalog sits inside the AI Assistant, which is in beta behind a waitlist, so the template library is the part every paying customer can reach today.

Sourceknowledge.apollo.io AI Assistant article and apollo.io/product/sales-engagementread 2026-09-04

Wide channel coverage inside one subscription, with several distinct entry points. Outbound runs through email sequencing, LinkedIn steps and a built-in dialer with call recording and transcription. Inbound arrives through form capture with enrichment and routing, website visitor identification at company and contact level, and buyer intent signals as triggers. So work starts on a prospect's behavior, not only on a rep's list build. Workflows fire the resulting actions without per-step rep involvement, and a Chrome extension carries prospecting into other web surfaces.

Sourceapollo.io/product/sales-engagement and knowledge.apollo.ioread 2026-10-01

The customer can bring their own model, and the route to it is gated. Apollo's help center documents connecting your own LLM key so that any LLM-powered action routes through the organization's existing subscription instead of consuming Apollo credits. It names OpenAI, Perplexity and Anthropic as providers a customer might already hold an enterprise subscription with. That is real provider choice over what powers the work, not just disclosure of a fixed dependency.

It is a restricted version: the connection lives inside the AI Assistant, which Apollo states is in beta and available only to selected users with a waitlist to join. No model list, per-agent selection or routing policy is published for customers who do not bring a key. Watch the terminology: several third-party sources use "bring your own key" to mean the Apollo API key rather than a model key, and those are different capabilities.

Sourceknowledge.apollo.io AI Assistant articleread 2026-09-04

A full developer surface plus a first-party MCP server that acts rather than reads. Apollo publishes developer documentation at docs.apollo.io with endpoints in OpenAPI and a machine-readable index at docs.apollo.io/llms.txt. API keys are created in-product at Settings, Integrations, API Keys, with per-endpoint scoping or a master key toggle.

The MCP server is Apollo's own, published in the apolloio GitHub organization and hosted at mcp.apollo.io over streamable HTTP. It authenticates through Apollo OAuth, so the server acts with the signed-in user's own Apollo permissions. It can execute, not only read: alongside people and company prospecting and bulk contact and account enrichment, it finds sequences and adds, removes or stops contacts in them.

Apollo also ships an optional plugin for Claude Code and Cowork bundling ready-made workflow skills over the same server, and names GitHub Copilot, VS Code, Cursor, Claude Code, Claude Desktop and Claude Cowork as supported clients. An integrations marketplace and a Chrome extension sit alongside. One limit on availability: API access depends on the plan, which Apollo states directly, directing buyers to upgrade or contact sales.

Sourcedocs.apollo.io, docs.apollo.io/docs/create-api-key and github.com/apolloio/apollo-mcp-pluginread 2026-09-04

A real experimentation mechanism, aimed at the outbound outcome rather than at the agent. Apollo ships A/B testing on sequence steps, so a customer runs one message against another and reads which performs. Deliverability scoring, automatic mailbox warmup and send-time optimization act on measured results rather than just reporting them. Nothing tests the agent or the automation.

No sandbox, dry run, preview of a workflow before it fires, scored evaluation of assistant output, regression check when a sequence or prompt changes, or comparison of one configuration against another is documented. So a change to what the platform sends is judged only after it has reached real prospects.

Sourceapollo.io/product/sales-engagementread 2026-09-04

The one candidate is an extension that surfaces data rather than operating software. Apollo's Chrome extension carries prospecting into other web surfaces, showing contact and company data alongside whatever the rep is already looking at and letting them act. That is assistance inside a page a person is driving.

Everything the platform does on its own runs through programmatic routes: CRM writes through native integrations, sending through connected Gmail and Microsoft mailboxes, calling through the built-in dialer, and external control through the documented REST API and the MCP server.

No headless browsing, virtual desktop, screen or click automation, or navigation of a third-party console by an agent appears anywhere on the product pages, help center or developer documentation.

Sourceapollo.io/product/sales-engagement and docs.apollo.ioread 2026-09-04