Tethra AI

Tethra's clearest idea is that service businesses don't have a prompting problem, they have a coordination problem. The homepage states the case bluntly: founders spend 60% of their time on coordination, operations live in four different tools, 3-4 hours a week go to manual tracking, and 44% of deals go cold from missed follow-ups. The product answer is a crew with named roles rather than a single chatbot with a system prompt. Sara takes inbound on WhatsApp, email, Slack or SMS, qualifies it, matches listings, and tags Omar for viewing prep; Omar books and preps docs and moves the pipeline stage; Layla queues the next touchpoint. The dashboard logs each of those moves with a timestamp, which is the part that actually lets a manager audit behaviour rather than trust it. The parts worth taking seriously are memory and handoff. Agent memory and context persistence across conversations is what separates a named crew from four disconnected bots, and the handoff mechanic — Omar acting unprompted on work Sara started — is the specific thing generic chatbot builders don't do, because they answer one conversation at a time. Training on your own files (listings.csv, fleet-data.xlsx, contracts.pdf, inventory.json) is the other real differentiator: answers come from your catalogue, not the open internet. The Custom Skills framework and REST/webhook connectors give a non-technical operator a way to reach into existing systems, and Salesforce and HubSpot are named explicitly. The honest weaknesses are on the outside of the product. Public documentation is thin — /docs, /changelog and /release-notes resolve to a sign-in screen, so there is no readable changelog, no setup guide, and no way to evaluate integration depth before creating an account. No underlying model is named anywhere on the site, which means you cannot reason about cost, latency or data handling from the outside. Android is still listed as coming soon, so an Android-first field team is stuck on the web app. There is a single paid tier at $99.99/year billed annually, so there is no graduated plan if you only want one agent or one channel, and no published enterprise security package for procurement. Where it fits: a Dubai or Riyadh brokerage, fleet operator or FM firm with high inbound volume spread across WhatsApp and email, an ops manager already acting as the human router, and a willingness to judge the product by running a real pipeline during the 14-day trial. Where it doesn't: teams that need to read docs and a changelog before they buy, developers who want to inspect orchestration logic, and anyone whose procurement requires a published compliance package. Compared with CrewAI or AutoGen, Tethra trades transparency for a working crew and a support line — a fair trade if you're an operator, a bad one if you're an engineer.