Multi-engine live intelligence collection
Fresh signals on every prospect, at processing time.
For each recipient, the engine queries multiple grounded-search providers in parallel — Perplexity, OpenAI-class engines, and more behind a provider-agnostic seam that lets us add new sources without re-engineering. Collection runs at processing time, not from a stale list bought six months ago. Results are cached and re-collected when the required signal set changes. The intelligence is current the moment the claim is minted.
Truth-grounded per-recipient claims
Every claim, specific to the individual recipient.
Every outreach message carries a claim that is specific to the individual recipient and grounded in a real, collected observation about them — not a merge-field swap. Each claim is verifiable: backed by a stored evidence reference, so the recipient (and the operator) can trace the claim to its source. Claims are atomic and composable: a claim is a single grounded fact; the email template composes the surrounding message. The truth-bearing unit stays clean and auditable.
Database-level truth gate
Architecturally enforced. Not procedural.
A claim cannot be selected for send unless it is verified and has a non-null evidence reference — enforced by the database via a generated column, not by application code that could be bypassed. A one-selected-claim-per-recipient constraint prevents conflicting sends. A composite foreign key ties selection to eligibility so an ineligible claim cannot be selected. These invariants are regression-tested by a schema guard that runs on every CI build. The integrity is structural.
Faithfulness guard on generated claims
Zero hallucination risk.
When the engine uses an LLM pass to phrase a claim, the output is validated against the grounded facts by a faithfulness guard before acceptance. Anything not faithfully entailed by the evidence is rejected, not sent. The rejection rate is visible on your dashboard — proof the guard is doing real work. Your brand stays trustworthy on first contact.
Claim-tuning studio
Sample, iterate, approve, mint at scale.
An operator-facing studio. Iterate on a small representative sample of prospects. Review the minted claims. Approve a configuration. Mint the full list under the locked configuration. A cost ledger tracks per-generation token cost so claim quality is tuned against real economics. A layered prompt architecture separates truth/format rules (locked) from style (editable) — you adjust tone without touching the rules that keep claims true.
Claim-type library (extensible)
Your ICP shifts. The engine extends.
Built today: AI-search visibility claims (you’re missing from ChatGPT/Perplexity answers where competitors appear) in two flavors — total absence and split-visibility. The claim-type library is architected so new types — built on entirely different evidence sources — slot into the same minting pipeline. On the roadmap: hiring signals, client ad-activity, funding events, product launches, technographic changes. The engine compounds as the library grows.
Managed sending portfolio
Every mailbox tracked, monitored, gated, rotated.
Every sending domain and mailbox tracked in a registry: provider, provisioning date, warmup start, computed ready date, daily send cap, health, campaign assignment. The registry auto-syncs from the sending platform — provisioned mailboxes appear automatically. A health monitor continuously evaluates each mailbox and auto-pauses degrading ones before reputation damage. Mailboxes are warmup-gated: the engine will not send live campaigns from an unready mailbox. Capacity is clamped to actual provider throttle — the engine never advertises more than the provider will deliver. Mailboxes rotate across campaigns from a ready pool of warm, healthy, available accounts.
Engagement + suppression handling
Closed loop between sending and deliverability.
Inbound engagement events (replies, bounces, unsubscribes, mailbox disconnects) ingest into a per-recipient engagement ledger and a global suppression list. Bounces and disconnects feed back into mailbox health, closing the loop between sending and deliverability monitoring. The recipient is never re-targeted across campaigns. The sending domain is never overcommitted.
Done-for-you campaigns Scale & Enterprise
Adella runs the engine. You get the meetings.
On Scale and Enterprise plans, the full outbound operation runs on Adella’s side. A dedicated outbound strategist tunes the claim-type configuration to your ICP, oversees the sending portfolio, runs the campaigns, triages replies, books meetings into your calendar, and reports to your CRM. You get the pipeline. We run the engine.