llm_chatgpt 0.4.0

OpenAI/ChatGPT backend implementation for LLM interactions. Provides streaming chat, embeddings, and tool calling via the OpenAI API.

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

0.4.0 - 2026-08-30

Added

• Streaming tool calls surface as they arrive on LLMChunkMessage.toolCallDeltas, instead of being withheld until the call finishes.
• extraHeaders on ChatGPTChatRepository and its builder — arbitrary headers on every request, including embeddings. Protocol headers and authorization always take precedence.

Fixed

• Mid-stream error events are raised instead of ignored. Parsed as an ordinary frame such an event has no choices and no usage, so it was skipped and the stream ended as a success carrying a truncated answer. The error's code is surfaced as statusCode so a mid-stream 429 or 503 is classified as retryable.
• Streamed tool call fragments are correlated by index rather than by id. Continuation fragments carry no id by design, and the previous fallback attributed them to the most recently seen call.
• Parallel tool calls stay separate when a server or proxy emits all of them with index: 0. A fragment whose id differs from the call open at that index now starts a new call instead of being merged into it.
• Non-streaming responses kept only the first tool call and silently discarded the rest, so a model that asked for parallel calls had all but one dropped — the caller ran one tool and answered as though that were the whole request. The same truncation applied when converting an assistant message back for replay, which rewrote history and taught the model its other calls never happened.
• GPTToolCall.fromJson tolerates a missing index instead of throwing into the converter's catch-all, where the whole event was silently dropped.

Changed

• The empty priming delta is no longer yielded.

0.3.2 - 2026-08-18

Changed

• Version bumped to 0.3.2 in lockstep with the other packages, and the llm_core constraint raised to ^0.3.2. No other changes to this package.

0.3.1 - 2026-08-18

Fixed

• The default HTTP client is now createLLMHttpClient(), which applies TimeoutConfig.connectionTimeout and bounds the connection pool.
• Documentation no longer claims Azure OpenAI compatibility. Azure needs a different URL layout and an api-key header; this package always sends $baseUrl/v1/chat/completions with a bearer token. Vision and reasoning-model support, both real, are now documented.

Changed

• Dependency floors raised: Dart SDK ^3.12.0 (was ^3.8.0), http ^1.6.0, lints ^6.1.0, test ^1.31.0.

0.3.0 - 2026-08-17

Added

• Reasoning-model support: per-model detection (gptIsReasoningModel), reasoningEffort/reasoningBudget mapped to a clamped reasoning_effort, and sampling params dropped where the API rejects them.
• Streaming usage via stream_options.include_usage, including LLMUsage.reasoningTokens.
• Reasoning deltas from OpenAI-compatible servers surface as chunk.message.thinking.

Fixed

• reasoningBudget no longer hardcodes reasoning_effort: 'low'.
• Streamed deltas without a role (everything after the first delta) were dropped when folding responses and broke tool loops; they now default to the assistant role.

0.2.0 - 2026-08-12

Added

• Provider-local functional integration coverage for manual tool calls, automatic tool execution, structured output, and low-cost live-test model defaults.
• Builder support for metrics, response cache, and rate limiting.
• Structured output support via StreamChatOptions.responseFormat:
  • JsonFormat() → response_format: {type: "json_object"} (native OpenAI API)
  • JsonSchemaFormat(name, schema, strict) → response_format: {type: "json_schema", json_schema: {name, strict, schema}} (native OpenAI API)
  • responseFormat is propagated through tool-call loops so format constraints are preserved across all turns
• Bumped llm_core dependency to ^0.2.0

Changed

• Requests now honor per-call timeout, retry, generation, cache, and metrics options.
• OpenAI live tests default to gpt-5.4-nano for chat and text-embedding-3-small for embeddings.

Fixed

• autoExecuteTools: false now exposes tool calls without executing them.
• HTTP client ownership and disposal are deterministic.

0.1.9 - 2026-02-28

Changed

• maxToolAttempts default increased from 25 to 90
• Bumped llm_core dependency to ^0.1.9

0.1.8 - 2026-02-26

Changed

• Bumped llm_core dependency to ^0.1.8 for tool calling stream visibility

0.1.7 - 2026-02-10

Added

• batchEmbed() implementation: delegates to existing batch-capable embed() (OpenAI embeddings API accepts array of strings).

0.1.6 - 2026-02-10

Fixed

• Ensured conversion from ChatGPT tool_calls to LLMToolCall always yields non-null, non-empty id values, synthesizing IDs should the OpenAI response omits them.
• Improved interoperability with llm_core's strict toolCallId validation for tool messages in tool-calling workflows.

0.1.5 - 2026-01-26

Added

• Builder pattern for ChatGPTChatRepository via ChatGPTChatRepositoryBuilder for complex configurations
• Support for StreamChatOptions in streamChat() method
• Support for chatResponse() method for non-streaming complete responses
• Support for RetryConfig and TimeoutConfig for advanced request configuration
• Input validation for model names and messages
• Improved stream parsing with GptStreamConverter

Changed

• streamChat() now accepts optional StreamChatOptions parameter
• Improved error handling and retry logic
• Enhanced documentation

0.1.0 - 2026-01-19

Added

• Initial release
• OpenAI/ChatGPT backend implementation for LLM interactions:
  • Streaming chat responses
  • Tool/function calling support
  • Embeddings
  • Compatible with Azure OpenAI
• Full compatibility with OpenAI API