A business agent shaped by what you teach it. Built around a structured playbook, tested in private, published to your website.
Product captures and source-backed interactions.
No invented dashboards. No live AI calls.
Customers ask about products, policies, prices and next steps. A useful agent needs a business identity, approved information and instructions for what to collect—and when a person should take over.
TACA turns that configuration into a playbook. Its application manages the surrounding lifecycle: account-scoped agents, provider provisioning, testing, publishing and conversation records.
Based on inspected application sourceNot an imaginary prompt editor. These are actual outputs from the inspected TACA compiler, generated offline with synthetic inputs. Change the configuration and inspect what changes.
These controls choose outputs generated by the original compiler with fixed synthetic inputs. This explanatory interface is not a screenshot of TACA’s management UI.
Collect required information in this order. Do not proceed past a required item until it is clear. Ask a clarifying question when needed. - Step 1. Product need. Ask: What product are you looking for? Required. - Step 2. Quantity. Ask: How many units do you need? Required. Validation: Whole number
Showing information collection for the English configuration. Product need is collected first. Explicit handover rule selected.
The compiler sorts the collection fields and turns required flags into explicit ordered instructions. This guides the agent; it is not a deterministic guarantee of model compliance.
The compiler can make identity, required questions and escalation explicit. Answer quality still needs evaluation. A generated instruction is not proof that a model will follow it on every turn.
Trace a published conversation, a private test or asynchronous provisioning. Inspect the components and the access boundary instead of watching an unexplained animation.
Inference is a managed cloud call. The surrounding business records remain part of TACA’s application.
The company archive identifies Kingsley as a TACA-based maintenance assistant for Kingstons Residential in Cardiff. The captured page describes request collection and prioritisation, with a conversational entry point.
Identity, business context, collection order and escalation are compiled into named sections. Guided defaults and custom configuration have defined precedence; advanced instructions have a separate path.
Configuration is recorded before a BullMQ job creates or reconciles provider resources. The worker performs a non-empty greeting smoke check before marking the agent active. That check proves basic readiness, not broad conversational quality.
The signed-in test path checks customer ownership. Public access requires a deployed identifier and runtime entitlement checks. Test and widget records have distinct channels; the inspected publish UI recommends testing, rather than enforcing a mandatory test pass.
The page separates captured interfaces, reproduced compiler output and an educational architecture. None of these are presented as live model performance or current whole-platform assurance.
| Material | Evidence basis | Boundary |
|---|---|---|
| Product images | Original assets published on the TACA product page | Captured test conversation and empty request form; not a new evaluation |
| Playbook lab | Outputs generated by inspected compiler source with synthetic inputs | New explanatory interface; no provider inference |
| System explorer | Chat, agent, conversation and worker implementation | Source-derived simulation; not live telemetry or provisioning |
| Kingsley | Company news record and authentic archived interface | Client publication approval pending; no invented downstream integrations |
Source review: 1 November 2026. The active implementation repository is toot-agent-cloud. Inspected status records are dated 2 October 2026; some README and roadmap state differs. The original pure playbook test file was reproduced locally, but the complete platform was not retested.
Public pricing and marketing have evolved relative to the inspected UX specification. No prices, accuracy scores, scale claims or historical test-count headline are used here. This corporate site’s Payload/PostgreSQL implementation is preserved separately.