Blueprint · relationship

05 · JobCache × AI

A vendor chatbot reaches JobCache through public search and evidence tools; JobCache sessions, optional accounts, capabilities, and charge targets remain product-owned.

Cell 2.3 · product · control · data · evidence
  • SOTA chat surface to Session capability: invoke with granted scope
  • Session capability to Public search + evidence tools: authorize
  • Public search + evidence tools to Grounded answer: ground response
invoke with granted scopeauthorizeground response
perspectivetarget

SOTA chat surface

The person converses in a provider-owned client; provider identity is not JobCache identity.

AI vendor
boundarytarget

Session capability

A JobCache session is the normal scope; an optional JobCache account may grant capabilities and identify a charge target.

JobCache App
servicemixed

Public search + evidence tools

Typed queries expose bounded roles, ads, freshness, provenance, and immutable result references.

JobCache App/Core
processtarget

Grounded answer

The chatbot cites JobCache evidence and returns saved work only to the authorized session.

AI vendor + JobCache
session context and evidencesearch and session toolsusage, policy, cost, and audit05 ↔ 08 · protocol, tools, and results
  1. 01 · perspective · target SOTA chat surface

    The person converses in a provider-owned client; provider identity is not JobCache identity.

    AI vendor
  2. 02 · boundary · target Session capability

    A JobCache session is the normal scope; an optional JobCache account may grant capabilities and identify a charge target.

    JobCache App
  3. 03 · service · mixed Public search + evidence tools

    Typed queries expose bounded roles, ads, freshness, provenance, and immutable result references.

    JobCache App/Core
  4. 04 · process · target Grounded answer

    The chatbot cites JobCache evidence and returns saved work only to the authorized session.

    AI vendor + JobCache

Flows

  • SOTA chat surface → Session capability invoke with granted scope
  • Session capability → Public search + evidence tools authorize
  • Public search + evidence tools → Grounded answer ground response
as built target / open Boundary arrows open the neighbouring module; ports stay machine-only

JobCache × AI

Outcome

Let a person use a SOTA chatbot to ask market questions, inspect roles and ads, refresh missing evidence, and save useful results in JobCache without turning the chatbot into a corpus operator.

Identity and scope

The normal product scope is an opaque JobCache session. JobCache may also have accounts, but an account is not the runtime address of every operation. It exists to grant capabilities and, where needed, identify a charge target. Session IDs and request IDs continue to address saved work, collaboration, delivery, and retries.

Provider identity remains separate. Linking ChatGPT or another vendor does not create or replace a JobCache account, session, capability, or billing relationship.

Tool surface

  • public role/ad search with freshness and uncertainty;
  • evidence and provenance lookup;
  • acquisition/refresh request with truthful pending or failure state;
  • immutable result/snapshot references;
  • session-scoped saved views and collaboration;
  • optional capability and charge-target introspection where JobCache accounts apply.

Every tool is narrower than the corresponding product authority. Corpus writes, source policy, suppression, billing changes, and operator recovery remain Management or product operations.

Invariants

  • Product/session identifiers never enter scraper evidence.
  • Account presence is optional for public use and never replaces session/request addressing.
  • AI answers cite JobCache evidence and preserve disagreement; they do not fabricate freshness.
  • A model cannot widen its session capability, select a different charge target, or invoke Ops.
  • Removing the AI integration leaves the public JobCache product complete.

Current state and completion evidence

JobCache already has UI/API/MCP perspectives, opaque shared sessions, public corpus search, and evidence foundations. The vendor-chat connector and assistant-specific tool profile are target. Acceptance requires UI/API/tool parity, session isolation, optional-account capability tests, charge-target authorization tests, evidence-grounded answers, pending/failure handling, and denial of corpus or Management commands.

Source: architecture/modules/jobcache-ai.md