Reply.io
Twelve year old multichannel sales engagement platform, founder led since 2014, serving more than three thousand companies with sequences across email, LinkedIn, WhatsApp, text and calls, a contact database of over one billion records with its live data supplier named on the homepage, and Jason, an agent that runs outbound in autopilot, copilot or approval mode. The commercial surface publishes the full ladder across four product lines: email volume plans from forty nine dollars monthly, an all channel plan at eighty nine, agent plans from five hundred a month priced on active contacts, agency plans per client, and per account channel add ons.
The platform is the index's clearest case of a split identity. The governance layer is among the deepest recorded, a trust page carrying a commitment not to train shared models on customer data, a dedicated artificial intelligence policy document, named privacy regimes, international transfer frameworks and a vulnerability disclosure route, alongside an agent interface with scoped permission keys, per tool safety annotations and documented confirmation gates.
The conduct layer sells the opposite: a LinkedIn automation product operating customer accounts action by action, purchased Google and Microsoft mailboxes available pre warmed, a peer network warmup, and a homepage statistic advertising infinite mailboxes rotated per campaign under an unpublished fair usage policy.
Capability Axes
Capability grades
17 of 17 axes rated · 11 graded A or B
The intelligence layer is heavy and specific, an agent with named autopilot, copilot and approval modes, research agents whose scope the user configures, generated sequences, response handling, personalization variables, and user selectable models, but the platform predates all of it and stands without it: remove the intelligence and a complete twelve year old sequencer with a billion record database and sending infrastructure remains. Under the removal test that is the definition of the second band, load bearing but not constitutive. The claims are credible, shipped in production with named modes rather than gestured at.
The first top band on this axis in the index, and the apparatus earns it. The agent runs in three named modes, autopilot, copilot and a shipped approval mode announced in product updates, and agencies choose between fully autonomous and approval operation per client.
The programmatic surface is where it becomes exceptional: one scoped permission key with granular scopes such as contacts read and sequences operate, per tool safety annotations that agent clients can gate on automatically, and documented procedures including dry runs, verification steps and confirmation gates installable as skills.
A mobile application exists specifically to review and approve generated replies and monitor mailbox and account health, and the trust page states operational oversight of how intelligence features are integrated. The oversight machinery is documented at the level of individual permissions and gates rather than asserted.
Model selection is exposed to the user in product, announced in the vendor's own updates, which discloses providers by construction, the trust page acknowledges that features rely on authorized third party intelligence providers with inputs processed solely to support the features used, a dedicated artificial intelligence features terms document exists at its own route, and the research agent's information gathering is user configurable, you decide what it should find and use. Off the top band: the providers are not named on the pages read, and no method or evaluation documentation was located.
A dozen named individuals with roles and companies carry quantified results, twenty to thirty percent monthly growth at a named agency with a linked case study, seven hours saved weekly per seller and half of closed deals following campaigns at a named travel company, fifteen thousand dollars in a first quarter at a named firm, reply rates above sixty percent at two named companies, and a customer stories library sits behind them, alongside nearly fifteen hundred reviews at a strong average on the major directory.
Off the top band: every percentage is self reported without method, the same homepage states the review count as fourteen hundred eighty in one section and twelve hundred plus in another, and headline claims of ninety eight percent delivery and tenfold opportunities carry no attribution.
The governance documents are better than the marketing behaves. The trust page names the European regime, its United Kingdom counterpart, the California statutes and the Canadian anti spam law, an acceptable use policy exists at its own route, and responsibility transfer is stated plainly, the customer must ensure lawful use.
Against that, the selling surface runs the other way: automated text messaging at scale is a priced add on with no consent framing anywhere read, and the blog publishes a piece teaching readers how to get bulk email addresses for free framed as a hack, address harvesting content on the same domain as the compliance commitments. The posture exists on a page marked to be from search engines while the growth content is indexed.
The trust page carries real substance: customer data ownership stated outright, a commitment that data is not sold, participation in the transatlantic data privacy framework with a link to the program registry where it can be checked, standard contractual clauses and the United Kingdom and Swiss mechanisms enumerated for transfers, a public data processing agreement route, retention and deletion practices acknowledged, and dedicated privacy contact routes.
Off the top band: the subprocessor list is available only on request rather than published, the policies themselves were not opened, and the trust page carrying all of this is marked to be excluded from search indexes.
The upstream supplier of the billion record live database is named and linked on the homepage itself, real time data powered by a named provider, the first named source provenance disclosure recorded in this session and exactly what this axis asks vendors to do.
Off the top band: one named supplier does not account for the whole database, no licensing terms or consent chain description is published, and the surrounding product surface includes an email finder extension and website visitor identification whose data sources are not similarly attributed.
The exposure is sold as a named product line at a printed price. LinkedIn automation at sixty nine dollars per account per month performs connection requests, messages, professional mail, attachments, voice messages, liking posts, following profiles and endorsing skills on the customer's behalf, agent plans include multiple accounts, and the agency product runs automation across many client accounts under one roof, all of it operating a platform whose terms prohibit automation, stated in the vendor's own feature lists.
The email side compounds it: the platform resells Google and Microsoft mailboxes, available pre warmed, specifically for rotation at volume, a use those providers' terms do not contemplate. The settled standard applies from the vendor's own words.
A dedicated responsible artificial intelligence section on the trust page makes commitments most of the category never writes down: customer data is not used to train generalized models for the benefit of other customers unless expressly agreed in writing, third party model providers process inputs solely to support the features used, generated content may contain inaccuracies and must be reviewed, and a separate artificial intelligence features terms document is maintained at its own route. Off the top band: the written unless clause leaves a negotiated door open, providers and their retention are unnamed, and the commitments live on a page excluded from search indexes.
Two product families cross from nondisclosure into manufactured authenticity, both in the vendor's own feature lists. Generated voice messages are sold on two channels, automated personalized voice messages on calls and artificial voicemails on the professional network described as hyper personalized, synthetic speech delivered in the highest trust format a recipient can receive, presented as a person's voice with no disclosure position anywhere.
Around them, the sending identity itself is rotated across unlimited purchased mailboxes so no recipient level reputation accumulates, and personalization variables are marketed for standing out and staying out of spam. The recipient is the party every one of these mechanisms is designed to be invisible to.
The deepest programmatic surface recorded in this index: a public documented interface with its own documentation site, a live model context protocol server whose tools carry per tool safety annotations, a command line client with installable procedural skills, four documented access layers to one platform, a developer community, an integration partners program, native bidirectional synchronization with the two major systems of record, the two major workflow platforms, browser extensions, a white label program, a public status page, a help centre and a training academy. The agent facing subdomain documents all of it as one surface behind one scoped key. This is the template the top band exists for.
Cloud software on unnamed infrastructure providers, with a public status page as a real operational transparency artifact. The trust page discloses that customer data may be processed in the United States and other countries where the vendor and its providers operate, with the recognized transfer mechanisms enumerated, which is a transfer disclosure rather than a residency offering: no hosting provider, region choice or residency option appears anywhere read.
The trust centre half of this axis is real: a maintained trust page describing encrypted transport, access controls, vulnerability management, backup and recovery, incident response with a notification commitment, a responsible disclosure route with a security contact, additional materials available under nondisclosure, and a public status page.
The certification half is absent: no independent attestation, audit or certification is named anywhere read, unusual for a twelve year old platform, and the trust page itself is marked to be excluded from search indexes, so the transparency exists but is not findable, the published versus retrievable pattern applied to the vendor's own trust surface.
The published ladder is unusually complete across four product lines: email volume plans from forty nine dollars monthly on annual terms with eleven printed contact tiers and the active contact unit defined in a footnote, an all channel plan at eighty nine, agent plans from five hundred monthly priced on active contact volume with annual and monthly figures both printed, agency plans at two hundred ten monthly and five hundred per client, channel add ons at sixty nine and twenty nine dollars per account, a free trial without a card, and a migration discount.
Two load bearing gaps hold it off the top: infinite mailboxes are subject to a fair usage policy that is not published, the unpriced ceiling on the core resource, and the credit and validation add on price values render blank to a machine reader, an interactive slider serving its numbers only to a browser.
Ownership of customer data is stated outright on the trust page, retention and deletion practices are acknowledged with deletion requests honored under applicable law, a data processing agreement is public, and the plan tables name a statistics export to comma separated files.
What is missing is the exit machinery itself: no full record export function, format commitment, post termination window or retention period is specified anywhere read, and the migration tooling and discount the vendor markets all point inward.
The settled rotation position applies in its most industrialized form, cited from the vendor's own homepage, which prints an infinity symbol as a product statistic: infinite mailboxes rotated per campaign, beside ninety eight percent delivery and fourteen days from purchase to first send.
The machinery is complete and openly architected for volume past filters: purchase Google or Microsoft mailboxes new or pre warmed, pre warmed domains arriving with three mailboxes already carrying manufactured sending history, automatic authentication record configuration, a peer to peer warmup network of real inboxes defended as no bots, and copy that says rotate them across sequences to send high volumes safely without overloading any single account.
The tooling underneath is genuinely real, per inbox reputation tracking, validation, authentication setup, which is precisely the pattern the position describes: real discipline instruments in service of a volume architecture, with the sole stated limit an unpublished fair usage policy.
Coverage is stated across multiple dimensions: six named use cases from outbound sales to public relations and link building, dedicated products for agencies with per client pricing, recruitment and startups, a white label program for resellers, more than three thousand companies across ten plus years, a toll free North American sales line, and international transfer machinery implying a global customer base.
The active contact tiers from one thousand to half a million give real scale banding. Off the top band: no geographic or language coverage statement, and no articulation of who the platform is wrong for.
What Changed
Material product, compliance, evidence and commercial changes at Reply.io, each verified against a live source and tagged to the capability axis it bears on. Funding rounds and awards are not product changes and are not logged.
Reply.io moved its AI SDR API surface into public beta, taking 39 endpoints from documented to callable. The set covers the full AI SDR sequence lifecycle including autopilot enable, disable and force start, pending approvals with bulk send, Live Data searches, sequence preview and personalization, offer generation, the Strategist and Inbox AI drafts. Documentation was restructured into API Reference, MCP, Skills and CLI sections alongside a new command line client at v0.3.0.
Pricing
What this vendor charges, what it commits to in writing, and where the bill can move. Figures the vendor publishes itself are labeled Vendor Published. Figures labeled Estimated come from other sources and the vendor has not confirmed them.
- ›The company's own pricing page says the entry plan starts from $59 per person a month and multichannel from $99. Its own listing on a big review site says $49 and $89. Both are the company's numbers and they do not match, so get your figure in writing.
- ›The multichannel plan does not actually include the other channels. Professional network automation is $69 a month per account on top, calls and texts are $29 a month on top, and extra mailboxes start at $20. One seller running all channels is reported to land near $187 a month.
- ›The AI agent is a separate product starting at $500 a month, not a plan you upgrade into. Nobody agrees on how it is charged, so if that is what you are buying, that is the conversation to have first.
- ›The best thing here is a definition. The company spells out that an active contact means one person you can send a first email to plus unlimited follow ups in a month. That lets you work out what your allowance actually buys, which most companies leave vague.
- ›There is a free tier and a 14 day trial with everything switched on and no card needed.
How the price works
What you are charged for, and what makes the bill go up.
Per user subscription for the core platform, with each non email channel sold as a separate add on and the autonomous agent sold as a separate product. The served pricing page states four starting figures: the email plan from $59, the multichannel plan from $99, the agency plan from $166 and the agent product from $500. Both an annual and a monthly control are present and the words billed annually appear near the figures, so the billing basis of the served numbers is not established by the document. The vendor's own listing on a major review platform, marked as supplied by the vendor, states $49 and $89 for the first two plans.
The metering unit is defined precisely and published: an active contact is a unique contact to whom one first step email and an unlimited number of follow ups may be sent in a month. Sending within that allowance is unlimited, so the meter is reach rather than volume. Plans carry differing mailbox allowances, with 5 and 10 mailboxes at lower tiers and unlimited mailboxes and unlimited active contacts at the top, and the vendor states that deliverability warm up is included with every mailbox.
Add ons are published with rates: professional network automation at $69 monthly per connected account, calling and messaging automation at $29 monthly, additional mailboxes from $20 monthly, and a further line from $39 monthly. The calling and messaging figure is labeled per account in one place and per user in another on the same page. Fifty live data credits are included monthly against a database the vendor states exceeds one billion contacts.
The agency plan is priced per workspace rather than per seat and covers unlimited users and unlimited client accounts from $166 monthly. The agent product starts from $500 monthly and third party reporting on its metering is contradictory, describing it variously as per agent, per contact and per credit, with per agent figures reported between $300 and $1,000 and no vendor surface resolving it.
A free tier exists alongside a fourteen day trial with full feature access and no card required. Independent reporting places monthly billing 20 to 30 percent above annual.
What the contract says about your data
What the vendor commits to in writing once your data is in the product.
Published but thin, and the commercial terms in the agreement are sharper than the privacy ones. A data processing agreement is referenced within the terms of service rather than published as a standalone reachable document, and the addresses I tried for a dedicated one and for a security page both returned nothing. A privacy policy is published carrying a data retention section and addressing additional rights available to residents of certain United States states under the Californian regime, with a deletion route stated. A trust and security surface is named in the site navigation.
Three terms in the agreement belong in a buyer's assessment because they allocate risk rather than describe practice. The vendor disclaims responsibility for loss, corruption, deletion or failure to store customer data, and for any cost of backing it up or restoring it, which for a platform holding sequence history and reply data places continuity squarely on the customer. Fees are billed in advance and are non refundable except where law requires or an order form says otherwise. And any refund, credit or accommodation the vendor does give is stated to be discretionary and non precedential, which forecloses arguing from a previous concession. The vendor also reserves the right to investigate suspected violations and suspend or terminate access.
What I could not establish across the routes tried: a sub processor list, any named certification or attestation, a retention period expressed as a duration, a residency or hosting statement, and a route to obtain a signed processing agreement. That is a partial retrieval rather than a confirmed absence and should be rechecked before it is treated as a finding.
The custody question is broad because the channel coverage is. A configured account holds mailbox access, an automated professional network connection, dialler and messaging activity, and records drawn from a database the vendor states exceeds a billion contacts.
Getting started
What it costs and what is included before the product is running.
Not published as a fee, and the trial is generous and clearly stated: fourteen days with full feature access including the contact database, integrations and the model features, with no card required. A free tier also exists beneath the paid plans, reported independently as carrying 200 emails and 200 data credits monthly with the browser extension and sequence generator.
What replaces an implementation fee here is a stack of published add ons, and they matter more to a first year budget than any setup charge would. Professional network automation is $69 monthly per connected account. Calling and messaging automation is $29 monthly, labeled per account in one place on the page and per user in another. Additional mailboxes start from $20 monthly. A further published line starts at $39 monthly. Fifty live data credits are included monthly, which independent analysis consistently describes as exhausting quickly, with top ups reported near $39 per 5,000 credits or roughly $0.05 to $0.10 per lookup, none of which was confirmed on a vendor surface.
One structural constraint is reported rather than published and should be confirmed: that a connected professional network account consumes a seat slot, so the network add on and the seat it attaches to are not independent purchases.
The agency plan starts from $166 monthly and covers unlimited client accounts with centralized management, which is a genuinely different unit from the per seat plans and is the one place the vendor prices by workspace rather than by person.
No minimum term, migration charge or professional services rate was located. The terms state that fees are billed in advance and are non refundable except where law requires or an order form provides otherwise.
What to watch for
Where this pricing can surprise a buyer who has not read it closely.
The vendor publishes two different entry prices for the same plan, on two surfaces it controls.
Its own pricing page states that the entry plan starts from $59 and the multichannel plan from $99. Its own vendor supplied listing on a major review platform, labeled as pricing provided by the vendor, states $49 and $89. Independent guides published across 2026 report $49 and $89 as the annual rates with monthly running 20 to 30 percent higher, which would reconcile the pair if the served figures are monthly. The page carries both an annual and a monthly control and the words billed annually sit near the figures, so which basis the served numbers belong to cannot be determined from the document. A buyer reading either surface has a defensible number and they differ by $10 per user per month on both tiers.
What the vendor does publish exceptionally well is its metering definition, and it is the best single sentence encountered in this tranche. An active contact is defined as a unique contact to whom one first step email and an unlimited number of follow ups may be sent per month. That converts an abstract allowance into something a buyer can calculate against a target list, and it makes explicit that follow ups are free while reach is metered. Most vendors in this category leave exactly that question open.
The add on structure is the finding a buyer would otherwise be blindsided by, and here the vendor is straightforward where its resellers are not. The multichannel plan does not include the non email channels in the way its name implies. Professional network automation is a published add on at $69 monthly per connected account, and calling with messaging is a published add on at $29 monthly, described in one place as per account and in another as per user on the same page. Extra mailboxes start from $20 monthly and a further line starts at $39 monthly. So a single seller running genuinely multichannel outreach stacks three or four charges, and independent analysis puts the realistic figure near $187 per user monthly against a headline in the high double digits.
The autonomous agent is a separate product starting from $500 monthly and is not a tier of the platform. Independent reporting on how it is metered is contradictory, variously per agent, per contact and per credit, with figures from $300 to $1,000 per agent and multi agent deployments reported in the thousands. Nothing published resolves it, and for a buyer evaluating this vendor for its agent rather than its sequencer, that is the entire question.
One correction to the public record belongs here. Two independent analyses state that this vendor's pricing page blocks automated access and present their figures as necessarily approximate for that reason. It does not. The page serves 705 kilobytes to a plain request, and the site's exclusion file blocks only demonstration and scheduling paths, not pricing. My own first attempt failed on a timeout and would have reproduced their error had I recorded it.
No dollar figure is recorded in the numeric field. The vendor states two different entry prices on two of its own surfaces and the served page does not establish which billing basis its figures use, so any single number recorded here would be presented with more confidence than the evidence supports.