PreviewDETERMINISTIC BOT CRAFT · PREVIEW

V2 Craft

Turn approved work definitions and quality feedback into dialog, prompt, knowledge and test changes. Evidence retrieval, Copilot analysis, human approval, atomic mutation, validation and conversation and regression testing safely accelerate deterministic-bot delivery and expansion.

Product flow

01V2 snapshot & evidence
02Plan, approval & mutation
03Validator, playground & trace
From input to a verifiable outcome

DESIGNED FOR

Designers and engineers who need to build CogInsight V2 chatbots and voicebots faster for sensitive financial and public-sector work while strictly preserving policy, approval, references and source structure

OUTCOMES

Start with customer change, not a feature list

Verified live-site capabilities and current product sources are reframed around the customer workflow and operating outcome.

01

Create the bot with its tests

Create dialogs plus normal, exception and boundary tests from approved completion criteria, preserving work-dialog-test traceability.

02

Models plan; V2 owns policy

Keep final conditions for transactions, identity checks and state changes in V2 Dialogs while blocking arbitrary model-authored code or JSON.

03

Validate new and existing work together

Send new-work tests with the operating regression baseline to the Xi Quality Studio Test Workspace to prove the existing service remains intact.

CAPABILITY SYSTEM

The capabilities that make the product work

These are product-level capabilities customers can adopt and operate—not isolated buttons or controls.

01

V2-preserving schema

Parse V2 exports while preserving source fields and reference structures without loss.

02

Hierarchical graph

Visualize dialog.then, node.then and node.body branches in their native V2 hierarchy.

03

Evidence-first Copilot

Plan changes only from evidence in official manuals, ProductKB, live catalogs and the current bot snapshot.

04

Feedback-guided remediation

Bind Test Workspace failures and stakeholder feedback to exact Bot and test revisions, then interactively propose prompt, dialog, workflow, DocStore and test remediation.

05

Revision-bound mutation

Reject writes from stale snapshots through expectedRevision and apply only approved atomic mutations.

06

V2 validator

Check internal and external references, IDs, hierarchy and expression complexity before save.

07

Runtime playground

Separate anonymous bot sessions from management-authorized registry and trace access while validating real utterances and execution paths.

OPERATING FLOW

How it works after adoption

  1. 01

    Fix the approved baseline and snapshot

    Confirm the approved work definition, target bot, V2 revision, catalog and operating regression baseline.

  2. 02

    Plan design and tests from evidence

    Use only grounded fields, nodes and variables to plan dialog and test changes together.

  3. 03

    Human approval and atomic mutation

    After the user approves the plan and impact, apply a change bound to expectedRevision.

  4. 04

    Analyze feedback and approve remediation

    Analyze failures and stakeholder comments into grounded changes, then have a person approve remediation scope.

  5. 05

    Validate, converse and regress

    Validate structure and real V2 conversations, then run the exact changed revision with the existing operating baseline.

TRUST & DEPLOYMENT

The operating environment and responsibility boundary are part of the product.

01

Independent V2 product

It is not a mode of Xi Craft and independently owns V2 schema, validators, catalog, evaluation and approval history.

02

Trusted proxy / BFF

Use an authentication-injecting trusted proxy in deployed environments and never persist management tokens in localStorage.

03

Separate test from deployment approval

Passing validators and conversation tests does not equal production approval; customer-specific security review remains separate.

VERIFIED SCOPE

Evidence and boundaries behind the public copy

  • V2-specific WorkState, schema, graph-core, validator and mutation contracts
  • No writes before plan approval and stale-snapshot rejection through expectedRevision
  • The Xi integration surface is limited to the legacybot reference contract for deployed and verified V2 bots