iBrain · powered by Claude

AI for cold email that actually does the work.

iBrain is built on Claude. Not a thin GPT wrapper. Not "AI agents" that are really just templates with names. Six concrete things it does — with examples you can read.

what iBrain does · 6 capabilities

Concrete tasks.
Concrete outputs.

01
Generate emails from a brief

Describe the prospect and your angle. iBrain returns 3 variations of subject + body, each with a different opener, all in your tone (learned from your previous campaigns).

// brief
First touch for a B2B SaaS CTO at a Series A. Soft pitch on cold email infra. Mention their recent post about scaling SMTP.
↓
// variation 2 of 3
subject: the SMTP problem you wrote about last week

Saw your post on rolling your own sending infra at Acme. The IMAP-idle race condition you described is the exact thing imail's reconnect logic was built to solve…
02
Improve copy you already wrote

Paste a draft. Get back: stronger opener, clearer CTA, shorter (or longer — your call), tone matched. iBrain shows you the diff.

// your draft
Hi {firstName}, I wanted to reach out to introduce our platform which helps companies like yours…
↓
// improved · "shorter, stronger opener"
− Hi, I wanted to reach out to introduce our platform…
+ Hi {firstName} — quick one. Your team posted about…
03
Classify every reply automatically

Every reply gets a label. You stop scrolling past 50 vacation messages to find one human reply.

// inbound reply
Sounds great, let's set up a call. Tuesday 2pm PT works on my end — send the invite.
↓
// iBrain · classification
{
  label: "interested",
  confidence: 0.94,
  signal: "explicit_meeting_intent",
  action: "pause_sequence + route_followup"
}
04
Score deals based on reply content

Reads the actual reply, not just intent labels. Updates the lead score in your CRM-style view automatically.

// 3 replies, this morning
A: Sounds great, let's set up a call → score 0.91
B: Maybe in Q3 → score 0.42
C: Take me off your list → score 0.02
↓
// crm-style view update
A · escalated to fast-follow
B · scheduled +90d, paused current
C · suppressed globally
05
Suggest the next best action per lead

Reads the thread, the lead's tone, your scheduled follow-ups, and recommends: send follow-up / book a meeting / pause / manual reply. You approve or skip. Two-click outbound.

// thread context
4 messages · last reply 6 days ago · status: warm but stalled
↓
// recommendation
action: "send a short check-in"
draft: "Hey — circling back since you mentioned Q2 timing was tight. Worth a 20-min walkthrough now that the dust has settled?"
confidence: 0.81
06
Knowledge-base-aware Q&A

Upload your product docs, sales playbook, FAQ. When a prospect asks "what about SOC 2," iBrain answers from your docs — not from generic training data.

// inbound question
What's your SOC 2 status?
↓
// answer · sourced from /docs/security.md (uploaded)
"We're SOC 2 Type II as of Feb 2026. Audit report available under NDA — happy to share when you're ready to evaluate."
why claude · not gpt

Two reasons that
matter to you, not us.

reason 01 · cost

Prompt caching cuts API cost ~60% on repeat queries.

Cold email workloads are repetitive: classify a reply, score a lead, suggest the next message. Same system prompt, same context shape. Anthropic's prompt caching hits hard on this pattern. We pass the savings on as a higher per-tier AI quota — that's the only reason it's in the marketing copy.

// average cost · per generation
without caching $0.012
with caching $0.0048 ↓ 60%
quota for you 2.4× higher
reason 02 · quality

Better on long context. Quantitatively.

Cold email conversations get long. Reply classification needs the full thread — not just the last message. Claude handles 200K-token contexts without quality dropoff. On our internal eval of 1,000 multi-turn threads, accuracy is +6.2pp over GPT-4o on threads > 8 messages.

// internal eval · multi-turn classification accuracy
threads ≤ 4 msgs gpt 92.1% claude 93.4% +1.3pp
threads 5–8 msgs gpt 88.2% claude 91.1% +2.9pp
threads > 8 msgs gpt 79.4% claude 85.6% +6.2pp

You'll never see "Claude" in the product UI. You'll just see it work. We mention it here because it's the kind of architectural choice that — three months in — you might want to know was made on purpose.

honest limits

AI quota per tier.
Because surprise overage bills are a tax.

A "generation" is one of: generate email, improve email, suggest reply, classify reply, score lead. You'll see remaining quota in the AI tab. We email at 80% and 95%. Heavy users — agencies running 50 client workspaces — should plan for Apex.

TierMonthly generationsWhat "1 generation" meansOverage policy
Spark100any of: generate · improve · classify · suggest · scorePause iBrain, keep sending
Surge1,000same$0.02 / gen overage, opt-in
Velocity2,500same$0.015 / gen overage, opt-in
Apexunlimitedsameno overage

AI that does the work, not just the talking.

See iBrain — start free trial