--- id: CON-PFM-LLM-005 name: Jazz AI Conversation feature: FEAT-PFM-LLM-07 product_area: platform module: llm revision: 1 status: draft product_intent: proposed implementation_evidence: partially-observed source_sha: react-console: baf6dae382d6f41dce976c7f848667ccc37e076f definition: >- A Jazz AI conversation is the staff-facing history of exchanges with the teammate, optionally grouped in a project or attached to a query or directory. owned_by: MOD-PFM-LLM synonyms: - agent conversation - chat not_to_be_confused_with: - 'A metered request, which is an accounting unit rather than a staff chat.' - 'The query workspace, which owns the draft being refined.' facets: - name: Message meaning: 'A user or assistant message carries text and a timestamp, with applied-change marking where relevant.' used_for: Presenting the exchange and its effect on work. - name: Project meaning: A named project with a description groups conversations and supplies general-reply context. used_for: Keeping an initiative’s discussions together. - name: Scope meaning: A query or directory identity attaches the exchange to its saved record or draft. used_for: Continuing assistance beside the correct work. - name: Applied change and undo meaning: Applied replies identify changes and support reversing the latest edit. used_for: Reviewing and undoing conversational edits. - name: Archive meaning: Archive hides an exchange from active lists without deleting it. used_for: Tidying history while preserving restoration. - name: Title meaning: A default or first-message title can be renamed to identify an exchange. used_for: Finding an earlier discussion. relationships: - to: CON-PFM-LLM-001 statement: Query refinement calls the metered interpreter; the prototype’s general reply follows a separate path. provenance: observed source: 'react-console:lib/queries/client.ts:interpret' lifecycle: - state: New meaning: The conversation has not received its first message. how_it_changes: The first message can supply its title. - state: Active meaning: The conversation appears in the active list and accepts messages through its supported scope. how_it_changes: Archive or deletion changes its availability. - state: Archived meaning: The conversation is retained outside the active list. how_it_changes: Restore or scoped reuse can make it active again. acted_on_by: - CAP-PFM-LLM-019 - CAP-PFM-LLM-020 visible_to: - ACT-PFM-032 evidence: - >- react-console at baf6dae382d6f41dce976c7f848667ccc37e076f: lib/agent/types.ts, store.ts, schemas.ts, client.ts; components/agent/*; components/console/agent-chat.tsx; app/api/agent/**; components/queries/query-workspace.tsx; components/record-explorer/use-explorer-assistant.ts. supersedes: [] --- # Jazz AI Conversation ## What It Is A Jazz AI conversation is the exchange staff see when they ask the console teammate for help. It contains messages, a title and recent activity, and may belong to a project or to the query or directory being edited. The full Jazz AI page supports browsing and organizing conversations. The header panel brings assistance alongside ordinary work, where a registered scope can connect replies to the page’s criteria or filters. The current implementation has the shape of a reusable history but stores it in shared console-server memory under a sample owner. Seeded examples appear in that store. Changes disappear when the process restarts, and owner checks do not provide actual person or workspace isolation. A product manager should therefore understand the concept and its intended value without mistaking the prototype store for durable customer conversation records. ## What It Is Not A conversation is not Platform’s metered request unit. Live query refinement uses Describe With AI and its Platform accounting path. General chat uses the console’s separate model call, while directory assistance uses local rules. A project is a conversation folder and context, not an application project record or a permission boundary. Archive preserves history and differs from confirmed deletion. ## How It Is Used Staff ask general planning questions, return to an earlier discussion or organize related exchanges into a project. When refining a query, they keep the conversation next to that query and inspect changes before saving. Applied changes can be undone latest-first, with a reversal note in the exchange. A conversation opened away from its scoped work directs staff back rather than letting general chat alter a detached query. Reply failure restores the draft so the person can retry. Good use treats answers as assistance and checks their relevance instead of assuming the general model can inspect live records. ## Its Facets **Message.** Text and time form the visible exchange. An applied marker connects an assistant reply with a change to work on screen. **Project.** A name and description group conversations and guide the general reply prompt. Deleting the folder leaves its conversations unfiled. **Scope.** Saved query identity or a draft key attaches the exchange to one query or directory. Reopening assistance can restore the same archived exchange. **Applied change and undo.** The latest edit can be reversed, retaining an explanation of the reversal in conversation history. **Archive.** Archived exchanges leave active lists and remain available for restoration rather than being removed. **Title.** The first user message supplies a short recognizable title after the default name, and staff can rename it. ## How It Relates The query editor owns query state and saving, while the directory owns its filters. Their assistant integration applies changes through those surfaces. Platform requests account for live query interpretation, but general conversation currently bypasses the request and allowance concepts. Conversation client writes go to the console agent service rather than the query service. ## What It Could Become The product should replace the shared sample-owner store with durable, correctly scoped records before customer use. It should also decide how general replies enter declared operations, metering, evidence and organization access controls. That work needs proof of privacy and restart persistence as well as proof of cost limits. The existing scope, archive and undo distinctions provide useful behavior to preserve when the ownership boundary changes.