Autonomous prospecting by Neura Agency
Neura Prospector is an autonomous B2B lead generation and outreach engine. It discovers companies, enriches the decision makers, qualifies them against your ideal customer profile with a recorded evidence trail, drafts the email and WhatsApp message, and sends both inside per-account capacity limits that it is not allowed to exceed.
Access is provisioned on request. Plans are Neura Lite, Neura Pro and Neura Ultimate. Sign in if your workspace is already live.
Signal pipeline
one lead / one day
Discover
Sources searched within budget
Qualify
ICP match, evidence recorded
Match
Offer and persona chosen
Draft
Email and WhatsApp written
Send
Queued inside both caps
Capacity, enforced before the send
stops at the ceiling
Per-account daily cap, enforced in SQL before the send is queued.
clamped
Stage 1 of 3: 20 messages a day until the number earns stage 2.
clamped
The problem
Every hand-off between a search tool, an enrichment tool, a spreadsheet and a sequencer is a place where a lead is lost, reached twice, or reached at the wrong moment.
Finding a thousand companies is a scraper. Turning them into a qualified, evidence-backed, reachable list is the work — and it is the work nobody has time to do properly.
A search tool, an enrichment tool, a sequencer, a spreadsheet, an AI subscription and a WhatsApp client. Every hand-off between them is a place where a lead is lost or reached twice.
Domains get burned and numbers get banned when automation ignores its own limits. Most sequence tools will happily send the message that costs you the sending identity.
How it works
The engine runs on a fixed daily cadence rather than on someone remembering to press a button. A plan is written at 08:00, the run starts at 09:00, and a sweep every five minutes handles follow-ups, warm-up evaluation and retries.
The 08:00 daily plan reads every running campaign, checks how much provider capacity is left, and writes the day’s channel allocations. Then the 09:00 run enqueues one strategy job per campaign, which searches your configured sources for new companies.
Records are normalised and de-duplicated, then scored against your ideal customer profile with the evidence kept. Anything confidently outside the profile is rejected. Anything uncertain moves to the human review queue instead of being sent.
Each lead is matched to your offer, given a persona, and written to twice — once for email, once for WhatsApp — using the templates and visual design you configured.
The engine picks the healthiest sending account with the most remaining capacity for the day, checks the per-campaign email and WhatsApp split, and claims each message exactly once. A send that has already gone out cannot be re-issued by a retry.
Webhooks return opens, clicks, bounces and replies. Follow-ups fall due on day +3 and day +5, and a sweep every 5 minutes publishes whatever is due, evaluates WhatsApp warm-up and retries anything that failed.
Features
Discovery, enrichment, qualification, matching, drafting, sending and tracking are one system here, not seven integrations held together by a spreadsheet.
You describe the market. The engine derives the regions, countries, industries and search terms from that description and runs them against Google search through Serper. A provider fallback chain — Apollo, then UpLead, RocketReach and Lusha, with Cognism and ZoomInfo available — keeps a campaign supplied when one source dries up.
Instagram, Google Maps and LinkedIn (through Evaboot folders) each keep their own search config: region, keywords, cities, countries, result limits and daily limits. Every campaign carries its own discovery budget, so one team cannot spend another team’s capacity.
Hunter, Clay and Apollo fill in email and phone. The decision maker’s name and role are attached, then the record is normalised deterministically — country, city, phone number — and de-duplicated idempotently, so re-running an import does not create a second copy of the same company.
Contacts that fail the send policy — or whose email or phone is already suppressed — are marked not eligible the moment they land. They are never queued, so a suppressed address cannot be reached even by accident.
Each lead is measured against your ideal customer profile and tiered Low, Medium or High. Every judgement records the evidence it used, under an explicit fact-trust hierarchy: the model is not allowed to invent numbers or statistics. Leads that clearly do not fit are marked Rejected - Non-ICP rather than quietly kept.
Low-confidence results are retried and re-qualified automatically. When every AI provider is exhausted and the answer is still uncertain, the lead moves to a Human Review Queue instead of being guessed at. Uncertainty becomes a task, not a silent assumption.
Write your product or offer in ordinary language and the engine derives the targeting from it. Leads are matched to that offer, then assigned a persona and a template per channel, so the email and the WhatsApp message speak in the right voice.
Messages are drafted per lead for both channels, with visual email design — banner, footer and call-to-action colours — controlled per template. The AI router tries Gemini, then OpenRouter, then Groq, under a hard spend cap so drafting cannot produce a surprise bill.
Level 1 sends nothing without a human approval. Level 2 auto-approves behind a canary. Level 3 runs unattended. A campaign starts at Level 1 and moves up only when you say so; the approvals and review queues stay visible at every level.
Provider webhooks return opens, clicks, bounces and replies. A detected reply updates the lead’s sales status, so the pipeline in the CRM reflects the conversation rather than a manual update someone forgot to make.
Bring existing leads in from a sheet, and optionally mirror the working set back to it every 30 minutes, keyed by lead id. Teams that already live in a spreadsheet get the engine without abandoning the sheet.
Each organization sees only its own data. API credentials are stored encrypted per organization, and an organization can bring its own AI keys. Operators get observability and issue tracking for the runs they are responsible for.
Safety and autonomy
Three independent gates stand in front of any real send, and a canary campaign climbs a fixed staircase. Every advance is conditioned on delivery health, and a bad bounce or failure rate pauses the campaign and quarantines the WhatsApp number automatically.
Canary ramp
sends per day / health-gated advance
Stage 1 opens at 10 sends a day. Each advance to 25, 50, 100 and finally 250 is checked against delivery health first. A campaign that starts failing does not get to keep its volume — it is paused, and the number behind it is quarantined.
Nothing leaves the building. The whole pipeline executes and you read exactly what would have gone out, and to whom.
A small, health-gated slice of real traffic. Each advance from stage to stage is checked before the next one opens.
One control stops sending everywhere. It is independent of the other two gates, so it works even if a campaign is misconfigured.
Idempotency keys plus a claim-based message handoff mean a retry or a replayed job cannot send the same message twice.
A bounce, a complaint, an unsubscribe or a STOP cancels that contact’s queued messages in the same transaction that records the event.
Every canary advance is conditioned on health. A bad bounce rate or failure rate pauses the campaign and quarantines the WhatsApp number automatically.
Audit events and AI decision logs record what the engine did, which provider answered, and why a lead was rejected or held for review.
Nothing sends until a person approves it. The default starting point for any new campaign.
Approved automatically, but only within the canary slice, and only while the health gates stay green.
The campaign runs unattended inside its own caps. The approvals and review queues remain available at every level.
Channels
Email and WhatsApp are not interchangeable pipes. Each has its own capacity model, its own health signals and its own failure mode, so the engine treats them separately and lets a campaign split its volume between them.
via Resend
via Evolution API
Cold outreach on WhatsApp through an unofficial API carries a real risk of a number being banned. Neura Prospector reduces that risk by keeping volume deliberately low, ramping in stages, watching delivery health, and quarantining a number automatically when it degrades. It does not remove the risk. If your volume cannot tolerate a lost number, run WhatsApp at a low share or leave it off.
Integrations
Named plainly, grouped by the job they do. Where a provider is part of a fallback chain, the chain is described rather than hidden.
Sources the engine searches to find companies.
Turns a company into a reachable decision maker.
Delivers the messages and returns delivery events.
Qualifies, matches and drafts. Routed in a fixed order, under a hard spend cap of zero.
Import, mirror and storage for the operators running the engine.
Who it is for
The engine is used by agencies and B2B teams running outbound at volume. These are the three shapes of team it fits best.
You are the person doing the outbound today, between everything else. The engine runs discovery, qualification and drafting on a daily schedule, so the pipeline does not depend on you finding an hour for it.
You need more volume without more risk to the domains and numbers you already own. Capacity is enforced per account in SQL, and the canary ramp means new volume has to earn its place.
Each client is an isolated organization with its own credentials, its own AI keys and its own sending identities. Nothing crosses between tenants, and you can see which runs are working.
Request access
There is no self-serve signup and no payment step. Tell us who you are and what you sell. If the fit is right, the team at Neura Agency creates a workspace and sends you credentials for the sign-in page.
Prefer email? Write to info@neuraagency.com.
Questions
Short, direct answers about access, sending identities, risk and how the engine differs from a sequence tool.
Request access with the form on this page. Requests are stored and reviewed by the platform team at Neura Agency. There is no self-serve signup and no payment step here: if your use case fits, a human provisions a workspace and sends you credentials, and you sign in from the sign-in page. Plans are Neura Lite, Neura Pro and Neura Ultimate, and what each one includes is agreed when access is provisioned.
Yes. You supply the sending identities. Email is sent through Resend, so you connect your own sending accounts, and each account keeps its own daily and hourly limit. WhatsApp is sent through Evolution API, so you pair your own numbers by QR code, one instance per number. Your credentials are stored encrypted, per organization.
The engine is built to prevent that, and it is honest about the residual risk. Three independent gates stand in front of any real send: a dry run, a canary, and a global kill switch. A canary campaign ramps at 10, 25, 50, 100 and then 250 sends a day, and every advance is conditioned on delivery health. A bad bounce or failure rate pauses the campaign and quarantines the WhatsApp number automatically. Per-account caps are enforced in SQL rather than in application code, so the limit holds even during a retry storm. The remaining risk is specific to WhatsApp: sending cold messages there through an unofficial API can still get a number banned. Keeping volume low and health-watched reduces that risk but does not remove it.
A sequence tool starts once you already have a list and a message. Neura Prospector starts a step earlier: it finds the companies, enriches them into named decision makers, qualifies each one against your ideal customer profile with a recorded evidence trail, matches the lead to your offer, drafts a personalised message per channel, and only then sends. It also runs email and WhatsApp from the same campaign, splits volume between them, and treats sending capacity as a first-class constraint instead of an afterthought.
Yes, and that is the default. Autonomy level 1 requires a person to approve a message before it is queued. Level 2 auto-approves only behind a canary, and level 3 runs unattended. Levels are earned, not assumed: a campaign starts at level 1 and you decide when it moves up. Separately, an uncertain qualification never reaches sending at all — it goes to a human review queue.
Four things: a plain-language description of your product or offer, an ideal customer profile, the sending identities you intend to use, and a description of the market you want to reach. The engine derives its regions, industries and search terms from what you write, so the quality of that description sets the quality of the leads.
The scheduling and worker layer is self-hosted. A daily plan runs at 08:00 and a daily run at 09:00, and a sweep every 5 minutes handles follow-ups that have fallen due, WhatsApp warm-up evaluation and retries. Each organization is isolated: it sees only its own leads, campaigns, credentials and logs.
Volume is set by your sending capacity, not by a promise. Every email account has a daily and an hourly limit that the engine honours, every WhatsApp number has a warm-up stage cap, and every campaign has its own discovery budget and daily cap. You raise capacity by adding healthy sending identities or by advancing canary stages, never by overriding a limit in the code.