Brew

Brew at `brew.new` is an AI-native email service provider rather than a generic writing assistant. It combines AI-generated email design and copy with audiences, automations, sending infrastructure, analytics, templates, domain management and developer interfaces. Teams can use Brew as the sending ESP or create content in Brew and export it into an existing platform.¹²

The product is deliberately accessible to both humans and software agents. Brew exposes a public API for contacts, domains, emails, templates, audiences and sends, and its documentation promotes MCP and agent workflows alongside ordinary browser use.³⁴

Brew's browser workflow combines chat-style generation with a visual canvas and manual editing. The documentation describes version history, test sends, scheduling and export options rather than treating the AI output as a final immutable artifact.⁵

The brand layer is central to the proposition. Brew can build a brand identity from a website and use it across generated emails, with reusable design baselines and multiple brand workspaces depending on plan. First-party material positions this as a way to produce email that follows existing visual and writing conventions without rebuilding templates manually for every campaign.¹⁶

The product also supports image/media generation and content analysis through multiple AI services. Brew's current subprocessor list names Anthropic, Google, OpenAI, xAI, AWS, Fal.ai and Cohere among the providers supporting AI features, with most model traffic routed through Vercel's AI Gateway.⁷

Because routing can span several providers, “Brew AI quality” is not reducible to one model version. It depends on the chosen feature, Brew's orchestration, brand context and upstream providers.

Brew is a real sending platform, not merely an exporter. Its documentation covers domain verification, contacts, audience validation, segmentation, automations, one-off sends, analytics and email-delivery operations.²

This matters because lifecycle email usually fails at the seams: copy is generated in one tool, HTML in another, audiences in an ESP, and performance data somewhere else. Brew's product design attempts to keep the generation and operational loops together.

The API extends those functions programmatically. Brew's current API v1 supports contacts, domains, emails, templates, audiences and sends, with keys bound to a brand and an official TypeScript SDK.³ That makes the platform potentially useful as infrastructure behind an internal agent, a product workflow or a custom marketing system rather than only through Brew's own UI.

Brew also explicitly documents running email operations from AI agents. Its integration materials mention ChatGPT, Claude, Grok, Cursor, Codex and MCP-compatible clients as possible control surfaces.⁴ These are connectivity claims rather than independent tests of autonomous campaign reliability; approvals and sending permissions still matter when an agent can affect real customer communication.

Brew supports a wide set of existing email platforms. The integration catalog includes Braze, Brevo, Customer.io, HubSpot, Iterable, Klaviyo, Mailchimp, Mailgun, Mailjet, OneSignal, Postmark and SendGrid, allowing generated content to move into established sending stacks.⁴

That makes adoption less binary. A team can use Brew for ideation, brand-aware generation and email construction while preserving its current contact database, deliverability infrastructure or analytics stack.

For larger organizations, this may be more realistic than replacing an entrenched ESP immediately. It also means Brew competes in two categories at once: as a standalone ESP and as an AI content/operations layer on top of other ESPs.

Brew's documentation explicitly requires consent-based sending. Its audience-hygiene guidance prohibits purchased lists and requires explicit opt-in, and the security/acceptable-use material requires documented consent, unsubscribe handling, accurate sender information and compliance with applicable email and privacy laws.⁸

That is important for an AI-native ESP because agentic generation can make producing and sending high volumes of email easier. A platform that automates campaigns still needs mechanisms and policies that discourage spam and preserve sender reputation.

Brew's current public monthly pricing is clear.⁶

Free costs $0 and includes 500 AI credits per month, up to 1,000 email sends per month, one sending domain and three brand workspaces. Brew's pricing page estimates roughly 50 emails on Sonnet for the stated credit allowance, but that is an illustrative consumption estimate rather than a guaranteed output count.

Starter costs $49 per month and raises the allowance to 5,000 AI credits and 10,000 sends, with three sending domains and five brands. It removes the Brew watermark, adds current premium model access, HTML download and priority support.

Growth costs $99 per month with 10,000 AI credits, 50,000 sends, five domains and ten brands. Pro costs $249 per month with 25,000 credits, 200,000 sends, ten domains and fifty brands. Enterprise pricing is custom.⁶

The pricing design is useful because both AI consumption and email volume are visible. Teams can reason separately about how much content they generate and how much they send.

There is still some uncertainty inherent in AI-credit systems. Brew's Terms say AI-token conversion rules can change with provider/model costs and quality and that excess usage can be rate-limited, queued, delayed or suspended until allowances reset.⁹ Buyers should therefore treat the listed credit quantity as the stable unit and any “number of emails” estimate as approximate.

The most important privacy fact in Brew's current terms is the tier distinction.

On Free, users grant Brew a license to use Inputs, Outputs and other Customer Data not only to operate the service but also to develop, test and improve Brew's models and algorithms, explicitly including training and retraining machine-learning systems. Brew says identifiable Customer Data will not be disclosed for those purposes and that its Privacy Policy and applicable law still apply.⁹

On paid tiers, Brew says Customer Data is processed to provide, maintain and secure the service and is not used to train generalized or shared models. Aggregated and de-identified data can still be used for analytics and quality assurance.⁹

Brew's security page states the distinction again in plainer language: limited samples from Free accounts may be used to train or fine-tune Brew-hosted models; paid plans are excluded from training, although Brew may evaluate limited samples from accounts to improve systems without model training.⁸

For individual experimentation, that may be an acceptable trade. For a company uploading proprietary strategy, customer-related content or confidential draft campaigns, paying for a tier with the more restrictive training terms may be important independently of volume limits.

Brew's GDPR documentation frames the customer as controller and Brew as processor when Brew handles recipient/customer data on the customer's behalf. It states that processing primarily occurs in the United States and uses Standard Contractual Clauses where required for international transfers.¹⁰

The security documentation says data is encrypted at rest with AES-256 and in transit with TLS 1.3 or higher, with row-level encryption for sensitive collections/tables. Brew says production backups use point-in-time recovery, are retained for 30 days and are globally replicated for resiliency.⁸

The current subprocessor list is extensive. Infrastructure includes Vercel, Cloudflare, Convex, MongoDB Atlas, Upstash, Turbopuffer, Clerk, Datadog and Braintrust; delivery-related services include Resend, Mailgun and Entri; AI services include Anthropic, Google, OpenAI, xAI, AWS, Fal.ai and Cohere; website/brand analysis can involve context.dev, Exa, Browserbase and Firecrawl.⁷

This transparency is a positive. It also means a security review should look beyond Brew itself because customer data can traverse multiple specialized providers depending on the feature used.

Brew's security page is unusually candid about maturity. It explicitly says the security program is in development and the company is working toward industry-standard certifications rather than claiming certifications it does not yet have.⁸

Current stated controls include least-privilege access, role-based access controls, vulnerability monitoring, defense in depth, AES-256 encryption at rest, TLS 1.3+ in transit, DDoS protections, a web application firewall, intrusion monitoring, code review, dependency scanning and CI/CD security controls.⁸

Multi-factor authentication is listed as a planned feature rather than a current universal control.⁸ That is a notable gap for a platform that can hold audience data and control outbound email campaigns.

Brew's DPA adds contractual data-protection commitments, including processor obligations and security/breach provisions.¹¹ Organizations with regulated data should still perform their own vendor assessment, particularly while Brew's formal certification program is still maturing.

Brew's largest weakness today is external validation.

Product Watch records Brew's launch as May 31, 2026 and describes it as an AI-native ESP that creates on-brand emails, campaigns and automations.¹² That is useful corroboration of existence and positioning, but it is not a performance test.

The practical implication is that evidence confidence remains low even though the product specification is unusually complete. We know far more about what Brew says it does than about how consistently real customers achieve good results with it.

Brew is most compelling for growth and lifecycle teams that already spend significant effort moving between AI tools, design systems and ESPs. It offers a credible route to collapsing that workflow into one environment while retaining export compatibility.

It is also interesting for startups and product teams that want email operations accessible to internal agents through an API or MCP, provided they build appropriate human approvals around real sends.

The Free plan is a useful functional trial, but organizations with confidential marketing data should read the tier-specific training terms before using it with real customer or proprietary information. Paid plans have materially different model-training terms.

Teams with strict compliance requirements should evaluate Brew's DPA, U.S.-centric processing and subprocessor list and may prefer to wait for the security certifications Brew says it is pursuing.