Swarmhit

A model layer sits across the product and the meters count senders and leads rather than inference.

What inference does here: writes personalised openers, drafts comments on a prospect's most recent post, scripts voice notes when a seller does not want to record their own, and interprets a plain language description of an audience into a database query. The protocol server extends that to an external assistant running whole campaigns conversationally.

What the customer pays for is different. Pricing is per connected sender at thirty nine dollars, or flat at fifty senders, or flat unlimited. Credits are consumed one per lead unlocked from the database. Not one unit of account measures model usage, and the model features appear in the entry tier feature list rather than as an upgrade.

The engine underneath is deterministic and it is what the second half of the business sells. Sender pools, rotation, daily caps, branching logic, health monitoring and safeguards are all rule based, and another product embedding this as infrastructure is buying that rather than the writing.

The plain language database search is the most interesting application, since translating a described audience into filters is work a person would otherwise do badly.

Ask what the openers are generated from and whether they can be disabled.

The highest consequence automated action in the product is the one that requires a signature, and the assistant integration carries the same commitment.

Publishing a comment under a real person's name on a prospect's public post is visible to that prospect's entire network and cannot be quietly retracted. It is the riskiest thing this platform can do, and it is the thing gated. The model drafts, and the interface presents approve, regenerate and reject controls against a pending draft. The vendor states plainly that nothing goes live until a human signs off.

That a vendor chose to gate the action with the largest blast radius, rather than the easiest one, indicates the control was designed rather than added for appearances.

The protocol server repeats it. An external assistant can find leads and launch campaigns in plain language, and the vendor states that the operator approves before anything sends. Assistant driven pipelines are precisely where unsupervised action would be simplest to allow, and it is not.

Around these: per sender daily caps shown numerically in the interface, sends spread evenly across the pool, live account health monitoring with a health percentage and limit tracking, automatic warm up, a published acceptable use policy, and role based permissions with single sign on above the entry tier.

The hundred plus safeguards claimed are not enumerated.

Ask whether comment approval can be disabled by an administrator.

Four model capabilities and no technology named behind any of them.

Two passes across the home page, the feature sections, the pricing tiers and the protocol server page located no model provider, no model name, no version, no statement of what prospect data is transmitted for inference, no training or retention position and no accuracy figure.

The voice note scripting is where the omission is sharpest. A model writes words which are then delivered as a spoken audio message to a prospect, presented as a personal note from the sender. That is generated content reaching a recipient in the most intimate available format, and nothing describes what produces it.

Comment drafting carries the same gap on a public surface, though the approval gate means a person reads it before it appears.

The protocol server compounds the question rather than answering it. Six named tools are exposed to external assistants, so a model outside this vendor's control can search prospects, create campaigns, add leads and send messages, and the boundary between what that assistant does and what the platform's own inference does is undescribed.

Ask which provider generates openers and voice note scripts.

Eight logos, two of which are unexpected enough to be the most informative thing here.

Among the named customers are two of the largest business contact data vendors in the world and a well known scraping platform. If accurate that says something specific: companies in the data business are embedding this as infrastructure rather than building network outreach themselves, which is a stronger signal about the developer half of the product than any testimonial would be. The remaining logos are smaller agencies and tools.

Beyond the logos the record is thin. Two passes located no case study, no named testimonial, no review count, no rating on any independent platform and no outcome figure attributable to a customer.

One trap worth recording for a future reader. The interface mockups on the home page display campaign figures of 1,240 invitations sent, thirty eight percent accepted and a 27.4 percent reply rate. Those are illustrative numbers inside a rendered product screenshot, not claims about customer performance, and a reader skimming would take them for results.

A status page operates and eight competitor comparison pages are published.

Ask which of the named logos use the platform and which embed the interface.

Real caps and published policies, on a channel that offers a recipient no exit.

What exists is genuine. Per sender daily limits are stated numerically in the interface at twenty two to twenty five invitations, with volume spread evenly across the pool rather than concentrated. Live health monitoring tracks each account against its limits. An acceptable use policy and a processing agreement are both published as separate documents.

The vendor also publishes guides on connection limits and on whether automation is safe, which is education about the constraints rather than marketing around them.

What is absent is anything protecting the person receiving the contact. Direct messaging on this network has no unsubscribe mechanism at all, so the only protection available is not contacting people who have not asked, and two passes located no suppression list, no do not contact handling and no exclusion mechanism of any kind.

No consent basis is stated for the database of more than a billion profiles that feeds the campaigns.

Every control described serves account survival. None serves the recipient, which on this channel is the gap that matters.

Ask whether a suppression or do not contact list exists.

A processing agreement published as standard and a custom one offered at the top tier.

The published set covers a processing agreement, an acceptable use policy, a privacy policy, a cookie policy and terms, each linked separately. The top tier adds a custom processing agreement alongside onboarding and a service level commitment, which indicates the vendor expects negotiated terms from larger buyers and has a process for them.

Access controls are real above the entry tier: role based permissions and single sign on, which for an agency operating fifty senders across many clients is the mechanism that keeps one client's conversations away from another's staff.

What two passes could not locate: a subprocessor list, a retention schedule, a deletion commitment or a residency statement.

The holdings deserve naming because one category is unusual. Alongside a database of more than a billion profiles and the conversation history in the consolidated inbox, the platform holds working access to connected network accounts, and in the case of rented senders it holds access to accounts belonging to people who are not the customer at all.

That third party custody is the question a processing agreement would need to address.

Ask how access to rented accounts is governed and by which agreement.

A billion profiles, refreshed weekly, and no stated origin.

That count is the largest claimed anywhere in this index, ahead of every dedicated data vendor built this session. It is searchable by describing an audience in plain language and unlocked at one credit per lead, and the vendor states it refreshes weekly, which is a specific cadence and the only concrete thing published about it.

Two passes located no supplier, no collection method, no consent basis, no accuracy claim and no verification standard.

The customer list makes the question more pointed rather than less. Two of the largest business contact data vendors and a scraping platform appear among the named logos. A reader can draw more than one inference from that, and none of them is settled by anything the vendor publishes.

Profile data assembled at that scale about individuals on a professional network, where those individuals created their profiles for a different purpose entirely, is the provenance question this axis exists for.

Weekly refresh implies continuous collection rather than a licensed static snapshot, which sharpens it further.

Ask where the billion profiles originate and what the weekly refresh collects.

Account rental, and it is the second instance recorded in this index.

The feature is stated plainly and shown in the interface: spread sends across your own profiles, and rent vetted accounts when you need more capacity, with a rented sender sitting in the rotation beside owned ones under the same daily cap. That means outreach leaves from a real person's professional identity which the customer does not own, to prospects who have no relationship with that person.

Everything else compounds it. Automated connection requests, messages, paid messages, profile visits, follows, likes and published comments, rotated across a sender pool with per account caps set below the thresholds that trigger review. Warm up touches exist to generate feed presence before the approach.

Every one of those actions is prohibited by the platform's user agreement, and the caps and rotation exist because it is enforced.

The vendor prints on every page that it is not associated with or endorsed by that network, which is honest and is simultaneously the admission that no permission exists.

What distinguishes this from the worst record in the index is the absence of evasion language. Nothing here is marketed as undetectable or as staying under the radar; the safeguards are framed as keeping accounts healthy.

Ask who owns the rented accounts and what they have consented to.

Two genuine approval gates, and custody of accounts belonging to people who are not the customer.

The safety controls are credited in full on the oversight axis and they are real: comments drafted then held for approval, rejection or regeneration, and an explicit commitment that assistant driven campaigns require approval before sending. For a model layer that writes publicly visible content under a real person's name, gating the publication step is the right control.

Stewardship documentation is reasonable: a processing agreement, an acceptable use policy, a status page, role based permissions and single sign on above entry, and a service level commitment at the top tier.

What is absent on the model side: no provider, no evaluation, no accuracy figure and no statement of what a drafted comment or scripted voice note may not say.

What is absent on the security side: no encryption statement, no access control description, no incident response process, no breach notification commitment, no named security contact, no vulnerability disclosure route and no certification.

The custody position is the sharpest concern. Operating rented accounts means holding working credentials or sessions for individuals outside the customer relationship entirely, and nothing describes how that access is secured or bounded.

Ask how rented account credentials are stored and who can use them.

A rented identity is the most complete form of the problem this axis measures.

When a message arrives from a rented account, the recipient sees a real named professional with a real profile and a real history. That person did not write the message, does not work for the company selling, and has let their identity be used for a fee. Everything the recipient can check is genuine and the conclusion they will draw from it is false. This is the second instance in the index.

Layered above it, several features exist specifically to manufacture familiarity. Automated likes on a prospect's recent post, automated follows and automated profile visits are grouped by the vendor as warm up touches with the stated purpose of showing up on their feed before you ask for anything. Each generates a notification suggesting a human took an interest.

Drafted comments appear publicly under a real name on the prospect's own post, visible to their whole network. The approval gate means a person read it first, which is a real control, and it discloses nothing to anyone who sees it.

Voice notes may be scripted by a model and delivered as personal audio.

Ask whether recipients of rented account outreach are ever told.

Half the business is selling the engine to other products, and the surfaces reflect that.

The interface runs on its own dedicated subdomain with campaigns, senders and leads exposed as resources, and the published example shows a campaign created with a sender pool and a daily cap as parameters. Webhooks carry named events for acceptance and reply. White label infrastructure and reporting let another product ship this capability under its own brand.

A protocol server has its own page listing six named tools an external assistant can call, covering lead listing, campaign creation, lead addition, message sending, inbox listing and statistics, with the vendor stating that every other platform action is available over the same connection. It is presented as working with three named assistants and any compliant client. That is the ninth such server recorded this sweep and the most thoroughly documented.

A dedicated page addressed to artificial intelligence assistants sits in the company section, the tenth deliberate optimisation of that kind found in this sweep.

Around them: a status page, workflow integrations, an affiliate programme, eight competitor comparison pages and a guide series aimed at developers building their own tools.

Ask what the interface rate limits are and how sender pools are allocated.

One infrastructure detail is visible in an example and the rest is unstated.

The published interface example names a sender pool with a regional prefix, which indicates senders are grouped and allocated by geography. That matters operationally on this channel, because an account suddenly operating from an unexpected country is a well known trigger for platform security review, so regional pooling is a deliberate design choice rather than an accident of naming.

A status page reports live and the platform is described as monitored around the clock.

What two passes could not locate: a hosting provider, a region for the platform itself, a data centre, a residency commitment, a tenancy or isolation model, an encryption statement, a backup position or a continuity plan. A service level commitment appears at the top tier without a stated figure.

The holdings make residency worth resolving: a database of more than a billion profiles, conversation history across every connected sender, and authenticated access to network accounts including rented ones belonging to third parties.

A European buyer would need the processing location before connecting anything.

Ask which regions sender pools are available in and where platform data resides.

A reasonable governance set for the size, with nothing independently attested.

Published: a processing agreement, an acceptable use policy, a privacy policy, a cookie policy and terms, each linked separately, with a status page alongside. Above the entry tier, single sign on and role based permissions. At the top tier, a custom processing agreement, dedicated onboarding and a service level commitment.

That is a more complete legal position than most vendors at this price point, and the tiering suggests the company has been through enterprise procurement at least once.

What is absent is verification. Two passes located no certification of any kind, no audit report, no examination, no penetration test, no trust portal, no dedicated security page, no encryption statement, no subprocessor list, no vulnerability disclosure route and no named security contact.

The custody position makes this consequential rather than routine. An agency on the fifty sender tier has connected fifty professional network accounts, some of which may be rented from people outside both the agency and its clients, and compromise would expose all of them simultaneously along with the conversations conducted through them.

Ask whether any independent security assessment exists and what the service level commitment guarantees.

Three published figures that let a buyer compute the crossover without a conversation.

Thirty nine dollars per sender monthly, nine hundred and ninety nine flat for fifty senders, two thousand four hundred and ninety nine flat for unlimited. Those numbers do the work a sales call usually exists to control: fifty senders at the per sender rate would cost one thousand nine hundred and fifty, so the agency tier is roughly half price at that volume, and a buyer can see immediately where switching pays. Publishing both the unit rate and the flat rate is what makes that arithmetic possible.

Yearly billing is stated as saving twenty six percent, an oddly precise figure that suggests it is calculated rather than rounded for marketing.

Each tier enumerates what it adds. The entry plan includes the interface, webhooks and the protocol server rather than reserving them for larger buyers, which for a platform selling itself to developers is the right decision. Higher tiers add analytics, single sign on, role based permissions, a success manager, a custom processing agreement and a service commitment.

Credits are explained at one per lead unlocked, and daily sending caps appear numerically in the interface.

A seven day trial requires no call.

The deduction: infrastructure pricing for the embeddable half sits behind a separate tab and was not captured, and the two upper tiers route to a demonstration booking.

Ask what the interface and white label infrastructure costs separately.

Three programmatic routes out, all included on the cheapest plan.

The interface exposes campaigns, senders and leads as resources, webhooks push acceptance and reply events outward in real time, and the protocol server allows an external assistant to list leads, read the inbox and pull statistics. All three appear in the entry tier feature list rather than being reserved for larger buyers, so a customer paying thirty nine dollars has the same extraction capability as one paying two and a half thousand.

That is the right decision and an uncommon one, since gating the interface is the standard way to raise switching costs.

What two passes could not locate: an export scope statement, a format, a deletion commitment, a retention position after cancellation or an account closure process.

One asset cannot travel by construction. Relationships built through rented accounts belong to the profile that sent them. Connections accepted, conversations held and any goodwill generated sit with a person the customer never employed, and end when the rental does.

Owned senders keep their own connections, which is the ordinary position on this channel.

Ask what happens to conversations conducted through rented accounts when the rental ends.

The account health mechanics are the most concrete recorded on this channel.

Daily caps appear as numbers in the interface, at twenty two and twenty five invitations per sender, with total volume spread evenly across the pool rather than concentrated on one account. A health panel displays a percentage, a within limits status, a count of invitations used against the daily allowance and a warmed up state per sender. Automatic warm up ramps new accounts. The vendor claims more than a hundred safeguards on every account and monitoring around the clock.

Warm up touches serve the same function, easing an account into visibility with passive actions before higher risk ones.

Publishing guides on connection limits and on whether automation is safe is education about the constraint rather than marketing around it, and it sits in contrast to another record this session that sold the same protections as making automation undetectable.

What is missing: the hundred plus safeguards are not enumerated anywhere, nothing describes what happens when an account is restricted despite them, and no measured data supports the approach.

The deeper issue is who carries the risk. Where a rented account is restricted, the consequence falls on its owner rather than on the customer whose campaign caused it.

Ask what the hundred safeguards actually are.

Four audiences, and the fourth is a different business from the other three.

Agencies, sales teams and recruiters each carry a dedicated page, and the pricing ladder maps to them cleanly: one sender for a founder or individual operator, fifty for an agency running many clients, unlimited for a revenue organisation that does not want to count. Recruiters are a genuine inclusion rather than an afterthought, since candidate outreach on this network uses identical mechanics.

The fourth audience is developers, and it is not a segment of the same product but a separate commercial motion. Selling white label infrastructure to other software companies means the buyer is a product team rather than a sales team, the contract is usage based rather than seat based, and the end users never know this vendor exists. Running both from one engine is a deliberate strategy and the logo set suggests it is working.

Eight competitor comparison pages indicate a crowded category the vendor is willing to be measured against directly.

What is absent: no geography, no industry vertical and no company size band beyond sender count, and no statement of which regions the profile database covers well.

Ask which regions the contact database covers most densely.