The AI Platform

The AI Platform (TAP) is a desktop workspace built around shared rooms where people and reusable AI specialists work together. Its core design combines persistent specialist agents, provider-agnostic model routing, a built-in editor and browser, workflow automation, and a Mini-App Builder that packages repeatable work into reusable team applications. The current site lists routing across OpenAI, Anthropic, Google, OpenRouter, Cloudflare, local models and private/self-hosted endpoints. ¹

TAP is differentiated less by a single model than by orchestration. Specialists can carry instructions, knowledge, tools and approval gates, while shared rooms expose work to multiple people rather than isolating it in one-user chat sessions. The platform also advertises citations, human review and audit trails for reusable workflows. Zephyr’s August 20 launch announcement says the product reached general availability after five months of early access. ¹²

The breadth is notable, but current performance claims are largely first-party. No exact-surface independent benchmark or substantive hands-on review was found, and the additional coverage located is mainly launch/context material. That limits how strongly claims about accuracy, speed, cost savings or production reliability can be validated. ³⁴

TAP has a usable free entry point and transparent self-serve team tiers. The current pricing page lists Free at $0, Starter at $10 per seat monthly or $8 annualized, Basic at $20 or $16 annualized with $25 per month of model credits, Pro at $60 or $48 annualized with $100 in credits, and Max at $200 or $160 annualized with $500 in credits. Enterprise pricing is negotiated. ⁵

This is competitive for teams that can use the bundled workspace, routing and automation layers rather than paying for several separate tools. The main economic caveat is model usage: included credits vary materially by tier and usage above allowances may continue at a markup, so total cost depends on model mix and workload. ⁵

The evidence base is asymmetric. First-party documentation is unusually complete for a young product: Zephyr publishes pricing, privacy, security, acceptable-use and service terms, and the site has a frequent changelog. The privacy policy states that customer workspace content belongs to the customer and describes processing across workspaces, model routing, specialists, mini-apps and related services. The Terms also warn that model outputs can be incomplete, incorrect or unsafe and describe organizational administration of permissions and providers. ⁶⁷⁸⁹

The security page provides a responsible-disclosure policy and defines covered TAP surfaces, but this review did not verify a TAP-specific SOC 2 or ISO certificate. Zephyr Cloud separately advertises SOC 2 compliance; that adjacent product claim should not be silently transferred to TAP without exact-surface evidence. ⁷

Independent evidence remains the weak point. No current exact-surface hands-on review or eligible numeric user-rating corpus was found. SmarToolbox describes the same collaboration, routing and mini-app architecture but has no reader ratings, while TipRanks' launch summary is derivative/contextual rather than independent performance evidence. ³⁴

TAP is best suited to software teams, AI-native teams and other small-to-enterprise groups that want multiple people and agents working inside a common operational surface. The strongest use cases are shared human-agent channels, reusable specialist agents, model routing, AI-assisted coding/browser work, and repeatable workflows that can be packaged as mini apps. ¹

It is a weaker fit for buyers who primarily want a mature, independently validated enterprise platform, a simple single-user assistant, or a product whose reliability and value are already supported by a deep third-party review corpus. Teams with high model consumption should also model usage costs rather than judging the subscription price alone. ⁵