flutter_ai_client 0.1.5

Provider-agnostic chat controller for flutter_ai. Wraps any LlmProvider in a Listenable UseChatController with optimistic send, cancellation, regeneration, and frame-batched streaming — without imposi [...]

Changelog

0.1.5

• Turn-sequence guard: a late event from a cancelled stream can no longer mutate the conversation or leak onto the events stream after a new turn starts.
• maxBranches (default 20) caps retained regenerations so a long chat can't grow without bound.
• tokenBudget: stop the agent loop once cumulative usage exceeds the budget (a cost ceiling on top of maxSteps).

0.1.4

• Thread management: ChatThread, a ChatThreadStore (list/delete on top of ChatStore), autoTitle(conversation), and an InMemoryChatThreadStore for demos/tests — enough to drive a multi-conversation sidebar.

0.1.3

• totalUsage getter on UseChatController: summed AiUsage across the conversation (feed an AiContextMeter or estimate cost).

0.1.2

• Agent loop: pass onToolCalls (and optional maxSteps, default 8) to UseChatController and it becomes an automatic agent — when a model turn ends with unanswered tool calls it runs the executor, feeds the results back, and re-prompts until there are no pending calls or maxSteps model calls have run. Without onToolCalls behavior is unchanged (the host drives tools manually via addToolResults). Cancellation/stop aborts the loop mid-flight.

0.1.1

• editMessage(id, text) / editLastUserMessage(text): edit a sent user message (keeping attachments), discard everything after it, and re-run from that point — starting a fresh branch set. Closes the previously dead "edit" affordance in AiMessageActions.
• Persistence seam: a ChatStore interface (load/save) plus an attachStore(controller, store, id) helper that debounce-auto-saves the conversation once each turn settles. History is still in memory by default; this makes saving/restoring a thread a few lines. AiConversation is already JSON-serializable, so a store is just encode/decode around your storage.

0.1.0

Initial release.

• UseChatController — a ChangeNotifier wrapping any LlmProvider:
  • optimistic, synchronous user-message append
  • sendText / submit / stop / regenerate / clear
  • live model/provider switching (setProvider, setOptions, setTools)
  • coalesced, injectable notification scheduling (frame-batched streaming)
  • raw events stream escape hatch
• ChatStatus (idle / submitted / streaming / error).
• Exposes stackTrace alongside error so failures can be reported with full context.
• A fatal (message-scoped) StreamErrorEvent tears down the active turn so a misbehaving provider can't keep mutating the conversation after a fatal error; tool-scoped errors remain non-fatal.
• Re-exports flutter_ai_core.