Nova Recruiter

Over 100 ATS platforms sync two ways with Nova Recruiter, and outreach goes out through the customer's mail and LinkedIn accounts. The FAQ states "two-way sync with 100+ ATS platforms, each custom-built by Nova, plus Nova's own native ATS & CRM if you don't already have one", and Nova's comparison pages name Greenhouse, Lever and Ashby.

Outreach sends across LinkedIn, Email, InMail and the Nova network in a single sequence, a sequence can go out "under a hiring manager's name and inbox instead of the recruiter's", candidates sourced and contacted "sync straight into the pipeline", and on agency plans migration from the customer's existing ATS is performed. Positive replies push candidates onward automatically. Those are authenticated writes into a mail account, a LinkedIn seat and a third-party ATS.

A third ecosystem arrives through MCP: Claude, ChatGPT, Gemini, Slack and Cursor can drive searches, shortlists and campaigns, so Slack joins as a surface without a bespoke connector. Focusing on the recruiting stack is a choice of market, not a limit on reach.

Sourcenovarecruiter.ai homepage FAQ, /pricing and /blog/nova-recruiter-vs-hireezread 2026-09-12

Four named agents run sourcing end to end, from search through outreach to handling replies, with the recruiter setting the flow and an outside assistant able to drive it. The pricing comparison lists "4 agents: filter, shortlist, message, respond", and the homepage describes "AI agents that automate filtering, selecting, messaging and replying" for a stated 98 percent reduction in sourcing time. On the agency all-in-one tier the set extends to "AI screening calls and interviews" and AI candidate reports built from the recruiter's interview notes.

The sequence runs end to end and the return leg is automated: a search is screened and ranked, a shortlist is built, personalized messages are generated per candidate, follow-ups fire on a schedule across channels, a response agent reads incoming replies and drafts answers, and a positive reply pushes the candidate into the pipeline.

The flow is customer-authored at named join points: any channel can lead, the recruiter sets the days between touchpoints, messages are previewed before sending, and the sending identity can be switched to a hiring manager. Participants are distinct: the recruiter who defines the search and approves, the four agents, the hiring manager who can own the outreach, collaborators on free seats in agency workspaces, and the candidate. An outside orchestrator can drive it too: through MCP an external assistant iterates the search, pushes candidates into campaigns and pauses, resumes or A/B tests them.

Sourcenovarecruiter.ai homepage and /pricingread 2026-09-12

A maintained database of over 800 million profiles, ranked by trained scores, is what Nova searches, alongside 25,000 pre vetted candidates and the customer's own synced pipeline. "Every plan searches the full 800M+ profile database with merit-based ranking", drawn from 30+ sources, standing between runs and far past any context window. Retrieval is semantic rather than keyed: a recruiter can "describe the role like you'd explain it to a colleague, and AI screens and ranks matches against your criteria", with a filter builder alongside, returning a typical 100 to 500 qualified results.

A trained ranking layer sits on the profile database: Talent Score is built on "150,000+ assessments" from Nova's recruiter network, Match Score measures fit to one specific search, and a blended Relevance score is sortable, three readable rankings over the same corpus. A second, assessed corpus is named: "25,000+ pre-vetted candidates from Nova's own network, already assessed once", searchable alongside the public database.

The customer's own estate is in scope too: two-way sync with 100+ ATS platforms reads the existing pipeline, and on agency and enterprise plans Nova's native ATS and CRM holds the customer's candidates, contacts and mandates in the same system that is searched. It is not a document retrieval engine with citations. Nothing states outright that the scoring agent grounds its judgment on the synced ATS records rather than only deduplicating against them.

Sourcenovarecruiter.ai homepage FAQ and /pricingread 2026-09-12

Every AI personalized message is shown to the recruiter before anything is sent, and Nova's privacy policy commits that no candidate is contacted without a human decision. The product FAQ states that multi-channel sequences run across LinkedIn, Email, InMail and Nova "with AI-personalized messages previewed before anything sends". Nothing reaches a candidate until a person has seen the text that will reach them: a hold on the one irreversible thing this product does, not a posture.

The commitment is also made in the legal layer: the privacy policy states that "candidates are never automatically contacted or evaluated without a human-in-the-loop decision by a recruiter, and Nova maintains audit trails to evidence such human oversight", with all final decisions taken by human recruiters who "retain full control over whether to contact, shortlist or hire". Nova positions its scoring as outside Article 22 GDPR on that basis, so the gate supports its own compliance claim.

Recruiters also define criteria and review shortlists. The site is marketed as "Hire exceptional talent, while you sleep", and both hold together: the person approves the sequence and its messages, after which follow-ups fire on the schedule the recruiter set. The privacy policy wording comes from a search index copy; the preview gate on the homepage stands on its own.

Sourcenovarecruiter.ai homepage FAQ, with /legal-texts/nova-recruiter-privacy-policy as corroborationread 2026-09-12

Single sign on and a security review come with enterprise plans, but Nova holds no security certification yet. SSO is documented twice: the enterprise section of the pricing page reads "Enterprise-grade from day one: SSO, security review, and a DPA included", and the FAQ adds that "enterprise customers go through our standard security review with SSO support".

What is missing is a held attestation. Nova says in three places that ISO 27001 is in progress, and no SOC 2, ISO or CSA certificate is claimed as held on the homepage security block, the pricing page or the legal center.

Nova also publishes GDPR compliance, with public data processed under legitimate interest and instant candidate opt out, and states alignment with the EU AI Act on transparency and risk classification.

It names a Data Protection Officer appointed under Articles 37 to 39, signs a DPA with every customer that requires one at no extra cost, and encrypts data in transit and at rest. None of these stands in for an attestation.

Sourcenovarecruiter.ai/pricing and homepage security blockread 2026-09-12

Each candidate's ranking comes with its reasons, and every reply is kept in one inbox, but no record of what the agents did is documented. Three scores are published and independently sortable: Talent Score as a quality signal built on 150,000+ assessments, Match Score as fit to one specific search, and a blended Relevance score, so a recruiter can see which signal put a candidate where. Nova's course material describes an agent that "shortlists the best fits with reasoning you can audit".

Outbound and inbound are retained in one place, since "every reply lands in one unified inbox", with the sending identity recorded per sequence, and campaigns can be paused, resumed and A/B tested, which implies per-campaign state a reader can inspect. Reply-rate reporting is the headline operational measure, and the privacy policy states that "Nova maintains audit trails to evidence such human oversight" of automated processing.

That audit trail has no documented surface: nothing names a run log, an action history, an export, a retention period, or an operator view of what the four agents did, such as which candidates the filter agent discarded and on what score, when the message agent sent and under whose identity, or how the response agent classified a reply. A score explanation answers why this candidate; an audit trail answers what the agent did, and only the first is documented as a surface. Nova's help center is not covered here.

Sourcenovarecruiter.ai homepage FAQ, /pricing and the Nova Recruiter privacy policyread 2026-09-12

What persists in Nova is the talent database and the customer's own records, not agent memory. Nova maintains a persistent talent database that lets recruiters reuse candidates over time, but that is the corpus the search agent queries, which is knowledge rather than memory.

The other candidates are the same shape: Nova's native ATS and CRM holds candidates, contacts, mandates and pipelines, the customer's application data; saved lists and shortlists are customer artifacts; the unified inbox's conversation history is a message store; and credits, roll-over balances and campaign state are account bookkeeping.

No named memory layer, retained preference set the agents reapply across searches, decision traces carried between runs, stated scope or lifetime for anything the agents hold, or customer control to inspect, edit, export or purge agent state appears anywhere.

Talent Score is trained once on 150,000+ assessments and applied globally, a model rather than a per-customer store, and its learning is absorbed rather than retained. Running Nova from Claude through MCP keeps the conversation context in the calling assistant, not in Nova. The pricing page itemizes the platform down to credit mechanics, seat counts and roll-over terms and names no memory component.

Sourcenovarecruiter.ai/pricing and homepage FAQread 2026-09-12

All data sits in EU based cloud infrastructure with no cross border transfers, and no other region or environment is offered. Nova states it in a dedicated tile on its security block, "Data stored exclusively in secure, EU-based cloud infrastructure with no cross-border transfers", and repeats it in the FAQ answer to "Where is my data stored?" One geography, stated by the vendor, with transfers ruled out rather than merely unmentioned.

No selection surface exists. The pricing page itemizes three commercial models line by line (self-serve tiers with credit counts and seat limits, agency licenses with seat minimums and contract lengths, enterprise credit volumes with roll-over terms) and carries no region, residency or hosting row at any tier, not even in the enterprise block where SSO, security review and DPA are the listed differentiators.

Nothing names a second region, a customer-controlled environment or a point at which a buyer chooses. Self-hosting, on-premise and a private customer-controlled deployment are not offered. A single region is a real position for a vendor selling GDPR compliance against US competitors, but a buyer needing US or APAC residency has no documented option, and a buyer needing EU residency has no choice to make.

Sourcenovarecruiter.ai homepage security block and /pricingread 2026-09-12

Every paid plan includes four ready made agents, but they are stages of one sourcing workflow rather than separate agents a customer picks among.

The pricing comparison lists "4 agents: filter, shortlist, message, respond", included on every paid tier rather than sold as an add-on, and the agency all-in-one tier adds screening-call and interview agents and AI candidate reports.

Reusable artifacts sit alongside them: multi-channel sequence structures composed once and reused, AI-personalized message generation per candidate, unlimited saved lists and shortlists, and client portals and mandate templates on the agency ATS.

Filter feeds shortlist, which feeds message, which feeds respond: a customer does not adopt the response agent instead of the filter agent, and nothing is individually installable or removable. There is no catalog, library page, marketplace or third-party or community contribution path, so what a customer can adopt is one product with four stages.

Sourcenovarecruiter.ai/pricing and homepageread 2026-09-12

Outreach runs across LinkedIn, email, InMail and Nova's own network on a schedule the recruiter sets, the agents act on incoming replies, and the agency tier adds screening calls.

Multi-channel sequences run across "LinkedIn, Email, InMail, and Nova", listed as contact channels on the pricing comparison as "LinkedIn + Email + InMail + Nova Network", and LinkedIn messages and connection requests are named separately on Nova's comparison pages.

Voice is present: the agency all-in-one tier ships "AI screening calls and interviews", and its agent list reads filter, shortlist, write outreach, reply to candidates, "and run screening calls and interviews", a synchronous fifth modality.

Schedules are customer-set and explicit: "Any channel can lead, whether LinkedIn, Email, InMail, or Nova, and you set the days between each touchpoint yourself", with automated follow-up reminders across every channel.

Events drive the return leg: "every reply lands in one unified inbox", a response agent reads incoming replies and drafts answers, and a positive reply pushes the candidate onward into the ATS. Identity is configurable per send, since the same sequence can go out under a hiring manager's name and inbox, and through MCP, Slack and any MCP client can launch and manage campaigns. Events, schedules and multiple channels are all documented in depth.

Sourcenovarecruiter.ai/pricing and homepage FAQread 2026-09-12

Nova Recruiter describes its models as its own, names no provider, and offers no choice of model. The pricing page itemizes three models line by line, down to credits per candidate contacted, extra-credit rates, seat counts, roll-over terms and renewal caps, and carries no model, LLM or provider row at any tier, and the four tiles of the homepage security block name no provider either.

Nova says "our AI models and merit-based scoring meet the EU AI Act transparency and risk-classification requirements", and the ranking layer is a proprietary Talent Score trained on 150,000+ assessments: a stated model estate with no provider identified and no customer-facing choice over it. No model picker, admin setting, per-agent assignment or bring-your-own-key path appears anywhere.

Nova is marketed as usable from Claude, ChatGPT, Gemini or Cursor, but that is the customer's own assistant calling Nova through its MCP server. A recruiter choosing to drive Nova from Claude is not choosing the model that runs Nova's filtering, scoring or message generation.

Sourcenovarecruiter.ai homepage security block and /pricingread 2026-09-12

Through an MCP server, an outside assistant can run searches, build shortlists and launch outreach campaigns in Nova Recruiter, on paid plans. The FAQ states: "Connect Claude, ChatGPT, Gemini, Slack, or dev tools like Cursor, and run searches, shortlists, and outreach campaigns straight from chat."

Launching an outreach campaign from an outside assistant is an action taken in Nova by a caller Nova does not control, not just a read. The homepage repeats it as running searches, building lists and "launching outreach sequences" from any MCP-compatible assistant, and MCP has its own solutions page in the primary navigation.

The inbound direction is the product's position, not a side feature: Nova markets itself as usable without opening its own interface, a customer testimonial reads "I opened Nova Recruiter once and I've been running it from Claude Code ever since", and Nova's course site describes an agent plugged into Nova Recruiter with "14 MCP tools" iterating searches, pushing candidates into campaigns, pausing, resuming and A/B testing from a terminal.

MCP access starts on Starter at €49 and is "available on paid plans", so the free tier does not carry it. No separate public REST API reference, endpoint list, authentication guide or SDK is documented, and custom integrations via API are not supported by anything published.

Sourcenovarecruiter.ai homepage FAQ, /pricing and becomeanairecruiter.novatalent.comread 2026-09-12

Reply rates on the agents' outreach are measured and campaigns can be A/B tested, but there is no documented way to test a change before it reaches candidates.

Reply rate is Nova's headline measure, stated as "2.5x higher reply rates" on the homepage and "2–3x higher reply rates than single-channel outreach" on the pricing comparison, and campaigns can be A/B tested with pause and resume control per Nova's course material.

An A/B test between two sequences is a comparison, and reply rate reads on what the message agent produced rather than on the customer's hiring funnel. Every reply landing in a unified inbox and classified by the response agent gives the signal a place to land.

Nothing puts a change under test: no test environment, holdout, sandbox, versioning, evaluator or scored comparable result about agent behavior is documented, and nothing states that a changed search, sequence or message template is run against a known set before it reaches candidates.

The A/B capability is named in course material rather than on a product page, and no readable verdict format is described. Talent Score's 150,000+ assessments are training data for the ranking model, not a harness that gates a change, and the pricing page's business-case calculators model hours and fees saved, which measures the customer's recruiting operation.

Sourcenovarecruiter.ai/pricing, homepage and becomeanairecruiter.novatalent.comread 2026-09-12

Nova Recruiter sends LinkedIn messages and connection requests on a person's behalf, unattended, but does not say how. Sequences send LinkedIn messages and, per Nova's own comparison pages, connection requests, alongside InMail, and a sequence can go out "under a hiring manager's name and inbox instead of the recruiter's", so Nova operates a named person's LinkedIn identity on their behalf, unattended, on a schedule the recruiter sets.

Nova's own replacement comparison keeps "LinkedIn Sales Navigator, 5 seats", which "stays in your stack", so the customer retains a LinkedIn seat that Nova drives. The product advances the action, not a person clicking each step. Email and the Nova network are reached through integrations and messaging channels rather than a browser.

This is one modality, and its mechanism is undocumented. Nova does not state how the LinkedIn leg executes, whether through a sanctioned partner path, a session or an extension. No hosted or local browser, desktop session, remote or local computer control or RPA is claimed anywhere. That the LinkedIn actions happen is documented repeatedly; how is not.

Sourcenovarecruiter.ai/pricing, homepage FAQ and /blog/nova-recruiter-vs-hireezread 2026-09-12