Spiritt

The pitch that separates Spiritt from every code-generation assistant is the second half of the job. Cursor, Replit Agent and v0 can get you an app; Spiritt's demo walks a single brief into a deployed B2B sales intelligence product — Postgres with pgvector, Next.js 16 on Hono behind Kubernetes, a Claude Opus scoring agent with a tool loop, an outreach sequencer, an admin dashboard, Stripe billing with RBAC, end-to-end plus 1,000 RPS load tests, three-region deployment — and then promises to stay on it, monitoring sessions, auto-deploying fixes and sending a weekly summary. That maintenance loop is what a non-technical founder actually lacks, and it is also the part you cannot verify from a homepage. What the documentation and changelog establish is a fast shipping cadence. Version 0.8 on Oct 3, 2026 put Kay 1.5 across all four tiers — Kay Nano for speed, Kay the recommended default, Kay Pro the intelligence sweet spot, Kay Max reserved for the hardest work — with lower prices on all four. 0.8.1 then improved voice generation, multilingual dubbing and music; 0.8.2 added motion video cloning that rebuilds a reference clip scene by scene as an editable project, starting on Kay 1.5 Pro; 0.8.3 added inline message editing, so a typo or a wrong turn no longer means restarting the thread. Add agent-purchased phone numbers and coordinated call fleets (0.7.15), full voice conversation with the same tools and memory, and a managed API catalogue where you pay only for successful calls in Spiritt credits, and you have an unusually wide surface for a young product. Where we stay cautious. The scope is enormous, and breadth plus autonomy is a reliability question, not just a feature list. Nothing we reached publishes service levels — the ~1.5 hours in the demo is marketing copy, not an SLA — and the price of all that scope moves with credit consumption across Kay tiers, media generation and paid API calls, which you should model before committing a production workload. There is also nothing in the material we saw describing SOC 2 or HIPAA coverage, so treat regulated and payments-critical systems as out of scope for now, and keep human review on anything that touches money or customer data. Where it fits: a founder or small team with a concrete brief, a non-critical first target and a tolerance for reviewing agent output. Start on Kay and step up to Pro only when Kay stalls, as the vendor itself advises. Where it doesn't: engineering teams that need auditable, deterministic change control over production code, or anyone who wants to own every line.