Table of Contents

Migrating to Mythosia.AI 7

Mythosia.AI 7.0.0 is a breaking release. Upgrade the related packages together:

<PackageReference Include="Mythosia.AI" Version="7.0.0" />
<PackageReference Include="Mythosia.AI.Abstractions" Version="3.0.0" />
<PackageReference Include="Mythosia.AI.Providers.Alibaba" Version="2.0.0" />

Install only the packages your application uses. Mythosia.AI already brings in the matching Abstractions dependency; the explicit Abstractions reference is mainly for libraries that compile directly against IAIService or the shared models.

Breaking API replacements

Removed in v7 Replacement
AIService.GenerateImageAsync(...) Cast or depend on IImageGenerationService, then call GenerateImagesAsync(...)
AIService.GenerateImageUrlAsync(...) Read GeneratedImage.Url when supplied, or use the authoritative inline GeneratedImage.Data and MediaType
AIService.MaxMessageCount Configure ConversationPolicy; without one, the full active history is sent
ChatBlock.RemoveFunctionMessages() Keep call/result pairs intact, or explicitly clear and rebuild the conversation
AIService.ExtractFunctionCall(...) Override ExtractFunctionCalls(...) and return FunctionCallBatch
CompletionProtocol.ExtractFunctionCall(...) Override ExtractFunctionCalls(...) and return FunctionCallBatch
ProcessFunctionCallAsync(string, Dictionary<string, object>) Override ProcessFunctionCallAsync(FunctionCall); the base service schedules complete batches
GrokReasoning.Off Use Auto to omit the parameter or None for Grok 4.3; Grok 4.5 cannot disable reasoning

The unsupported Qwen image-method overrides were removed as part of the same image API migration. QwenService remains a chat-completion provider.

Image generation

Image generation is an optional provider capability and its model is independent from the service's chat model:

using Mythosia.AI.Models.Images;
using Mythosia.AI.Services;
using Mythosia.AI.Services.OpenAI;

IImageGenerationService images = new OpenAIService(apiKey, httpClient);
var result = await images.GenerateImagesAsync(new ImageGenerationRequest
{
    Prompt = "A clean architectural facade study",
    Count = 1,
    Size = "1024x1024"
});

GeneratedImage image = result.Images[0];
await File.WriteAllBytesAsync("facade.png", image.Data);

OpenAI supports generation plus reference/mask editing. Gemini supports generation and reference-image editing, requires Count = 1, and does not accept a separate mask.

Multiple function calls

Providers can now return multiple calls in one assistant turn. Sequential handler execution remains the default, so existing registration code keeps one-at-a-time behavior:

using Mythosia.AI.Models.Functions;

service.DefaultPolicy = new FunctionCallingPolicy
{
    ExecutionMode = FunctionExecutionMode.Sequential
};

Opt in to bounded parallel execution only for independent, thread-safe handlers:

service.DefaultPolicy = new FunctionCallingPolicy
{
    ExecutionMode = FunctionExecutionMode.Parallel,
    MaxConcurrency = 4
};

Handlers may finish out of order in parallel mode, but Mythosia returns results to the provider in the original call order. Custom providers must preserve FunctionCallBatch and FunctionCallResultBatch correlation data.

Removed model constants

Constants for retired, unavailable, or deprecated models were removed instead of being kept as obsolete aliases. This includes the old GPT-5 snapshots, GPT-5.2 Codex, GPT-4 Vision/4o-latest aliases, GPT-4.1 Nano, retired Claude Opus 4/4.1 snapshots, and unavailable Grok 3 Mini support. Select a current constant from AIModels rather than copying a retired wire ID into application code.

GPT-5.6 Pro is a request mode selected with Gpt5_6ReasoningMode.Pro; there is no separate gpt-5.6-pro model ID.

Conversation history

The hidden message-count window was removed. Without a ConversationPolicy, requests now send the complete active conversation. Applications that need bounded history should configure token-aware summarization or trimming explicitly.

Before publishing an application update

  1. Compile after replacing removed APIs and model constants.
  2. Exercise both streaming and non-streaming function flows if your application registers tools.
  3. Confirm whether each handler is safe for parallel execution before enabling it.
  4. Treat image bytes and GeneratedImage.MediaType as authoritative; hosted URLs are optional.
  5. Review the full Mythosia.AI 7.0 release notes and Abstractions 3.0 release notes.
Edit this page