Onramp logo
OnrampAI-Powered CRM
Core
Channels
Demo Tenant
Free Plan
Settings
Onramp logoOnramp

Onramp

AI SDR and inbox operating system with guardrails.

Onramp helps service businesses run outreach and inbox follow-up with a thread-centric, policy-governed system. Start approval-first, prove safety, then progressively increase autonomy.

Request AccessExplore DigiX VoiceSee What Works Now

Why teams get stuck

Outreach bottlenecks are rarely about sending one more campaign. The real drag is the inbox: triage, qualification, follow-up consistency, and compliance. Onramp addresses that with a controlled decision loop, not a freeform chatbot.

  • Reps lose time in repetitive back-and-forth.
  • Important replies get buried while low-value noise consumes attention.
  • Manual qualification delays handoff and booking.
  • Teams need explainability before trusting AI execution.

How Onramp works

Onramp runs an event-driven loop where thread history is the source of truth and AI execution is gated by policy and risk.

  1. 1Outbound campaign sends run through provider adapters and thread-aware metadata.
  2. 2Inbound replies land at /webhooks/postmark/inbound and route back to the correct thread.
  3. 3Thread + message history is persisted as the system-of-record context for each contact.
  4. 4An inbox agent creates structured actions (with confidence + reasoning) and drafts replies.
  5. 5Execution is policy-gated: auto-send for safe paths, approval for higher-risk paths.

What works now (MVP)

Inbound webhook intake + thread routing via /webhooks/postmark/inbound
Thread and message persistence (append-only timeline)
Agent action creation and draft workflows with reasoning visibility
Inbox + thread detail UI with approval/send behavior
Stage model with rationale shown in UI (new → engaged → qualified → booked → closed)
CSV campaign send workflow endpoints via /ui/campaign_sends*
Sequence endpoints available behind feature flag via /ui/sequences*

What's coming next

Roadmap focus is deliberate: deepen autonomy safely and harden operations before claiming full production-scale maturity.

  • Expand LLM-first reasoning coverage across all decision paths
  • Complete end-to-end suppression enforcement across every send path
  • Mature approval workflow controls and review ergonomics
  • Deliverability hardening: warmup, rate adaptation, and mailbox pooling
  • Analytics dashboard depth and operational visibility
  • Voice channel integration into the same policy-governed decision loop
  • Production hardening for auth/tenant isolation at larger scale

Safety, compliance, governance

  • Approval-first MVP default: no blind AI sending.
  • Autonomy is policy-driven (assistive → mixed → autonomous), not hardcoded.
  • All AI actions are logged and auditable in agent action history.
  • Thread-centric system of record supports explainability and rollback-friendly operations.
  • Suppression/compliance controls are built into the architecture and expanded over phases.

Today vs Future

Area
Today (MVP)
Future
Execution model
Assistive-first with human approval for sends.
Policy-gated conditional autonomy and fuller autonomous execution.
Reasoning depth
Structured rule-based and staged logic in production paths.
Broader LLM-first reasoning with stronger confidence orchestration.
Operations
MVP-safe workflow with core reliability and audit trails.
Deliverability + analytics hardening for sustained production scale.

Roll out autonomy responsibly

Start with approval-first guardrails, validate outcomes, and progressively unlock autonomy where confidence and policy allow.

Request AccessOpen Inbox