CogniLead docs.
The CogniLead runtime in eleven sections. Each is a separate page so you can deep-link the part that answers your evaluation question fastest — architecture, the GDPR posture, the MCP surface, the API contract. The full implementation document lives in the repo (IMPLEMENTATION.md).
Architecture
6 minA six-stage event loop with retry plus dead-letter on every queue. Every transition is Chainlog-anchored before the next worker touches the row.
Data model
7 minFour core tables: signals (raw, immutable, TTL-bound), leads (one per company × product), sends (one per lead × campaign × step), suppressions (hard gate, email or domain).
Ingestion workers
6 minFive worker families, each producing typed signal rows. Apify for LinkedIn / Indeed / jobs.ch, GitHub GraphQL for dep references, HN Algolia for Who is Hiring, Crunchbase for funding, BuiltWith for stack enrichment.
Core intersect
5 minSuppression check → dedup → Chainlog anchor. No lead proceeds to personalization without proof of ingestion timestamp and a suppression-clean status.
Personalization via PHI Gateway
7 minEvery LLM call is routed by the recipient jurisdiction, never by the sender. CH and EU recipients hit EU-resident models. Strict JSON output. If technical_hook_verified is false, the lead is dropped.
AgentGate policy gate
6 minAgentGate evaluates every dispatch against the outbound-gdpr risk pack. Hook verification, body length bounds, sender warmth thresholds, jurisdiction consistency, blocked sales language. Denials are durable.
Deliverability infrastructure
7 minFive to ten sender domains rotated on a steady cadence. Six weeks warming before promotion. SPF + DKIM + DMARC on every tier, BIMI on Scale. Postmaster reputation drives an automatic circuit breaker.
Resend Automations integration
5 minPipeline hands a lead.qualified event to Resend Automations with the full sequence pre-rendered. Each step calls back to AgentGate before dispatch to catch leads that went stale.
GDPR posture
8 minLegitimate-interest under GDPR Art. 6(1)(f) with a documented Legitimate Interest Assessment. Corporate inboxes only. RFC 8058 one-click unsubscribe. Retention defined per-table.
MCP server
6 minThe @cognilead/mcp-server exposes six tools to MCP-aware IDEs. Two free, four gated. Gated tools return 401 with an inline signup URL so a developer can authenticate without leaving Cursor or Claude Desktop.
Customer-facing API
7 minA narrow HTTP API. POST endpoints for signals, leads, campaigns, sends. GET and POST endpoints for the evidence pack. All authenticated with pk_live / sk_live key pairs.
The compliance plays the spec assumes you know.
The 2026 LIA playbook for B2B cold outbound
Seven elements your DPO will accept, plus the common objections and how to address each.
How to respond to a GDPR DPA inquiry in 30 days
A day-by-day plan for assembling the evidence pack, drafting the response, and getting sign-off.
Where CogniLead fits — and where it does not.
CogniLead vs Smartlead
Deliverability incumbent. We trade raw volume for an evidence pack and EU-resident inference.
CogniLead vs Instantly
Polish over compliance. Two different shapes of customer; here is how to pick.
CogniLead vs Lemlist
Templates vs signals. Per-seat vs per-send. The cases where each wins.
Phase 2 opens M3. The API surface is final, the SDKs are not.
Tell us which language you are calling from and we will prioritize that SDK first. TypeScript and Python are baseline.