Closely

Named CRM connectors plus the vendor's own documented route to anything else. The pricing page lists Salesforce, Pipedrive, HubSpot and GoHighLevel on every plan, with Gmail and Outlook over OAuth and custom SMTP for sending, and the action surface is LinkedIn, email and CRM writes.

The help center documents a webhooks integration the customer wires with their own URL, pushing contact data out of campaigns, My Network and the Inbox on tag and note events, with emails and phones included by default, and titles it for use with Webhooks by Zapier. That route lets the customer connect further systems on their own, so the list is not closed. The route runs one way: the vendor states there is no API and nothing pushes data into Closely.

Sourcecloselyhq.com/pricing and help.closelyhq.com webhooks articleread 2026-10-01

Real unattended multi-step execution inside a deterministic scheduler. Sequences chain profile visits, connection requests, follow-up messages, InMails, post engagement and email steps across every connected LinkedIn account, run in Closely's cloud rather than on the user's machine, advance or halt on reply detection and connection acceptance, and push outcomes to the CRM; behavior-based branching moves a prospect to email when a connection request goes unaccepted.

This is a sequencer spanning many accounts in which the AI is confined to drafting message copy, with no goal decomposition, no dynamic replanning, no branching the agent decides beyond configured conditions and no tool selection documented. The Sales AI Agents the vendor names research a lead's profile and company site and write the message; they do not choose what to do next.

Sourcecloselyhq.com/linkedin-automation and closelyhq.com/multichannel-campaignsread 2026-09-02

A large data asset, not a knowledge structure. Closely ships a B2B contact database, a cascading enrichment engine that falls back through sources until a verified work email is found, live LinkedIn enrichment for job titles, companies and industries, and phone lookup, and message generation is grounded in the prospect's LinkedIn profile and company website read at generation time. Inbound webhook payloads from LinkedIn Lead Sync forms feed personalization.

What is absent is any customer-owned grounding structure: no brand voice or messaging framework object, no ICP or persona model the agents read, no signal taxonomy, and no retrieval layer over the customer's documents or collateral. Volume of contact records is not depth of grounding, and the corpus belongs to the vendor and its data sources rather than to the customer.

Sourcecloselyhq.com/data-enrichment, closelyhq.com/email-finder-tool and closelyhq.com/business-contact-data-providerread 2026-09-02

Configurable guardrails that protect the sender rather than the recipient or the output. What ships is real: smart daily activity caps, randomized delays between actions, working-hours windows, a warm-up mode that ramps activity gradually, a dedicated residential proxy per account, and automatic pausing of a sequence when a prospect replies. Every one of those exists to keep automation beneath LinkedIn's detection thresholds and protect the customer's account from restriction.

None gates an agent decision, reviews generated copy before it sends, or protects the person receiving the message; human involvement happens in the unified inbox after messages have gone out. No approval workflow, review before sending, sign-off by role or policy engine is published. On the other side, the vendor publishes an anti-spam policy and an opt-out manual, which is a documented commitment about recipients even though it is not a mechanism in the product.

Sourcecloselyhq.com/linkedin-automation and closelyhq.com/anti-spam-policyread 2026-09-02

Legal documentation better than the certification position, which remains absent.

Strong: a full DPA with standard contractual clauses governing transfers out of the EEA, a contractual commitment to background investigate every employee or contractor before granting access to client personal data, post-termination retention only on an aggregated or de-identified basis, a privacy policy honoring GDPR Articles 12 to 23 with a documented one-month erasure window, Article 13 and 14 notice to EEA data subjects, and a published anti-spam policy.

No certification: the vendor's own blog states it undergoes audits aligned with ISO 27001 standards, which is alignment rather than certification, and the complete site footer carries only privacy, anti-spam, terms and cookies pages, with no security page, trust center, sub-processor list or penetration test evidence anywhere on the property.

The same blog article names AES-256 at rest and TLS 1.3 in transit in one passage and states the exact protocols are not disclosed in another, so the article contradicts itself. The structural point is unusual: Closely takes custody of the customer's authenticated LinkedIn session inside its own cloud, so an uncertified processor holds standing access to an identity the customer cannot easily rotate.

Sourcecloselyhq.com/legal/data-processing-agreement, closelyhq.com/pp and closelyhq.com/anti-spam-policyread 2026-09-02

Campaign visibility across accounts, no agent record. A unified dashboard reports per-sequence performance across every connected LinkedIn account, tracking connection acceptance, reply rates, InMail replies, email opens and clicks, meetings booked and step completion, and an agency can monitor multiple client accounts from one place without logging in and out.

Outbound webhooks push activity events into the customer's own systems in real time, so a customer who wants a durable trail can build one externally, though the record then lives in the customer's system rather than the vendor's. What is absent is anything agent-specific: no audit log, no record of what the AI generated against what a human edited, no action-level trail attributable to a user, and no retention policy on activity history. A buyer can see campaign outcomes and cannot reconstruct what the system did on their behalf or why, which matters because it sends under the customer's own LinkedIn identity.

Sourcecloselyhq.com/pricing and help.closelyhq.com webhooks articleread 2026-09-02

Conversation and sequence state persist; nothing the agent maintains does. The unified inbox keeps LinkedIn and email threads across every connected account, the agency dashboard keeps per client campaign state, and sequence state (step reached, connection accepted, reply received) persists and drives execution. Those are the product's records, which are not memory: no cross session agent memory, no learned preference carried between campaigns and no retention or forgetting control at the agent level, and message generation reads the prospect profile fresh each time.

Sourcecloselyhq.com/pricing and closelyhq.com/linkedin-automationread 2026-10-01

A fixed jurisdiction and nothing the customer selects. Control: vendor cloud only, with no region selection, no on premises option, no private VPC and no way to keep the LinkedIn session on infrastructure the customer controls; the cloud model is what lets sequences run when the user's machine is off. Jurisdiction: the operating entity in the terms of use is Esterson Limited in Cyprus, and the DPA governs transfers out of the EEA under standard contractual clauses, which implies EEA default processing. The location is disclosed, but the customer has no say in it.

Sourcecloselyhq.com/terms and closelyhq.com/legal/data-processing-agreementread 2026-10-01

Template infrastructure at the sequence and message level, no agent catalog. Reusable campaign sequences combine connection requests, messages, InMails and email steps; message templates carry smart variables that AI personalization then varies per prospect; and campaign patterns are offered by job title and segment.

For agencies there is a white-label layer with a master dashboard, per-client portals, custom branding and domain, an unusual packaging capability. Sequence templates are configuration for one automation engine rather than a catalog a buyer can browse for named agents or templates aimed at named jobs, and no marketplace, importable play library or community exchange exists.

Sourcecloselyhq.com/pricing and closelyhq.com/white-labelread 2026-09-02

Two channels, driven by sequence state. Documented and working: LinkedIn, covering connection requests, direct messages, InMails, profile visits and post engagement, and email, with an integrated finder and sender for hybrid campaigns.

Triggers are sequence state transitions, connection accepted, reply received, step elapsed, plus inbound webhook events from LinkedIn Lead Sync forms, and behavior-based rules such as moving to email when a connection request goes unaccepted within a set period. Intake and coverage stay narrow as a result.

Phone appears in a multichannel claim on the vendor's blog, but no product surface supports it, and the same blog carries other claims the product pages do not support.

Sourcecloselyhq.com/multichannel-campaigns and closelyhq.com/linkedin-automationread 2026-09-02

No provider named and no customer choice. AI is described only functionally: Sales AI Agents analyze a lead's profile and company website and craft tailored messages, and AI personalization is metered as a credit at two credits per use. No model selection, routing, bring-your-own-model, bring-your-own-key, or temperature or style control beyond message templates is published.

The provider is simply undisclosed rather than a deliberate vertical integration, and the distinction matters to a buyer: for a product whose entire AI surface generates outbound messages sent under the customer's own name and LinkedIn identity, an undisclosed model provider is a live procurement question.

Sourcecloselyhq.com/linkedin-automation and closelyhq.com/pricingread 2026-09-02

One-directional egress and nothing else. What exists: a webhooks integration documented in the help center that pushes information from Closely to the customer's system and, in the vendor's own words, does not support pushing data in the other direction, plus a Zapier listing reaching downstream apps and a Chrome extension for LinkedIn data extraction. A customer can react to Closely events and cannot drive Closely programmatically. The name invites confusion.

A December 2025 post on the vendor's own blog describes a Closely REST API with named endpoints, an authentication model and a rate limit, but the endpoints, resource names, authentication split, cursor pagination, fields parameter and 429 handling in that post match the published developer documentation of Close, a different company with a near-identical name whose CRM the same blog writes about elsewhere.

No Closely API documentation, developer portal or reference exists on any Closely surface, and the vendor's own support answer states there is no API, so the post describes another company's API rather than a Closely one.

Sourcehelp.closelyhq.com webhooks article, Closely support response and closelyhq.com site navigationread 2026-09-02

Outcome measurement without any way to test first. Closely ships detailed campaign analytics: connection acceptance and reply rates, InMail replies, email opens and clicks, step-level completion per sequence and across accounts, campaign timelines and team performance comparison, plus A/B testing of message variations across LinkedIn and email, which lets a customer compare one sequence against another empirically. That measures the outcome of the agent's work rather than the agent.

Nothing tests before sending: there is no sandbox or dry run mode, no evaluation harness, no output scoring, no regression suite and no way to check a change to message generation before it reaches real prospects. For a product that sends under the customer's own LinkedIn identity, where a bad message costs the account rather than a campaign, the absence of a dry run is the consequential gap.

Sourcecloselyhq.com/pricing and closelyhq.com/multichannel-campaignsread 2026-09-02

The automation works inside LinkedIn itself. Closely operates the customer's authenticated LinkedIn session to perform profile visits, connection requests, direct messages, InMails and post engagement, running each account on a dedicated residential proxy so the traffic presents as a real user, with warm-up mode and daily caps configured to mimic human behavior.

LinkedIn publishes no sanctioned interface for these actions at this scale, so the product necessarily operates the interface rather than calling an API, which is action inside software the vendor does not control. A Chrome extension, published and listed on the Chrome Web Store, additionally extracts data from LinkedIn pages the user is viewing. This is automation built for one site, with no general browser or computer control and no way to point it at another application. Whether it runs a headless browser or an emulated client is not documented.

Sourcecloselyhq.com/linkedin-automation and closelyhq.com/extensionread 2026-09-02