Added the exact token count to chats which offer tools

app/MindWork AI Studio/Chat/ChatThread.cs

@@ -401,8 +401,7 @@ public sealed record ChatThread
     /// asked, and no earlier answer ever stands in for it. Whatever came after an older report -- a
     /// message whose request was turned down, an answer which is still being written, an answer
     /// without a report of its own -- is missing from that report's number, and the estimate is
-    /// closer to the truth than a figure which leaves it out. Answers written while tools were
-    /// offered carry no report yet, so for them the estimate always takes over.<br/><br/>
+    /// closer to the truth than a figure which leaves it out.<br/><br/>
     ///
     /// A report also stops counting when the thread holds a different number of blocks than the
     /// request did, which means an earlier message was deleted, and when the next request goes to

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionStreamPart.cs

@@ -5,15 +5,20 @@ namespace AIStudio.Provider.OpenAI;
 /// </summary>
 /// <param name="TextDelta">The text this line carried, empty when it carried none.</param>
 /// <param name="Sources">The sources this line announced, empty when it announced none.</param>
-public readonly record struct ChatCompletionStreamPart(string TextDelta, IList<ISource> Sources)
+/// <param name="Usage">What the provider said the request cost, unknown on every line but the one which carries it.</param>
+public readonly record struct ChatCompletionStreamPart(string TextDelta, IList<ISource> Sources, TokenUsage Usage = default)
 {
     /// <summary>
     /// The part of a line which says nothing to the user, such as a fragment of a tool call.
     /// </summary>
     public static ChatCompletionStreamPart Nothing => new(string.Empty, []);
-    
+
     /// <summary>
     /// Whether this part has anything to show at all.
     /// </summary>
+    /// <remarks>
+    /// The usage is not part of that: it is nothing to show, and whether it is passed on at all is
+    /// the adapter's decision, which knows which round this is.
+    /// </remarks>
     public bool HasContent => this.TextDelta.Length > 0 || this.Sources.Count > 0;
 }

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionToolCallAccumulator.cs

@@ -60,10 +60,16 @@ public sealed class ChatCompletionToolCallAccumulator(Func<ServerSentEvent, ILis
         // list either way.
         //
         var sources = readSources?.Invoke(serverSentEvent) ?? [];
-        
+
+        //
+        // The usage arrives on a line without choices at most providers, and next to the last
+        // piece of text at some. Read before the choices are looked at, it is not lost in either.
+        //
+        var usage = line?.Usage?.ToTokenUsage() ?? TokenUsage.UNKNOWN;
+
         var delta = line?.Choices?.FirstOrDefault()?.Delta;
         if (delta is null)
-            return WithSources(string.Empty, sources);
+            return WithSources(string.Empty, sources, usage);
 
         this.hasReadAnything = true;
         
@@ -80,10 +86,10 @@ public sealed class ChatCompletionToolCallAccumulator(Func<ServerSentEvent, ILis
 
         var textDelta = delta.Content;
         if (textDelta.Length is 0)
-            return WithSources(string.Empty, sources);
+            return WithSources(string.Empty, sources, usage);
 
         this.text.Append(textDelta);
-        return new ChatCompletionStreamPart(textDelta, sources);
+        return new ChatCompletionStreamPart(textDelta, sources, usage);
     }
     
     /// <summary>
@@ -183,10 +189,10 @@ public sealed class ChatCompletionToolCallAccumulator(Func<ServerSentEvent, ILis
     private static string? Coalesce(string? value) => string.IsNullOrWhiteSpace(value) ? null : value;
     
     /// <summary>
-    /// A part for a line which brought sources but no text, or nothing at all.
+    /// A part for a line which brought sources or a usage but no text, or nothing at all.
     /// </summary>
-    private static ChatCompletionStreamPart WithSources(string text, IList<ISource> sources)
-        => sources.Count is 0 ? ChatCompletionStreamPart.Nothing : new ChatCompletionStreamPart(text, sources);
+    private static ChatCompletionStreamPart WithSources(string text, IList<ISource> sources, TokenUsage usage)
+        => sources.Count is 0 && !usage.IsKnown ? ChatCompletionStreamPart.Nothing : new ChatCompletionStreamPart(text, sources, usage);
 
     /// <summary>
     /// One tool call while its fragments are still arriving.

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionToolCallingAdapter.cs

@@ -54,6 +54,16 @@ public sealed class ChatCompletionToolCallingAdapter<TRequest>(
             ParallelToolCalls = requestDtoBase.Tools is null ? null : false,
         };
 
+        //
+        // Only the first round passes on what its request cost. Its prompt is the conversation up
+        // to the question, which is exactly what the next question will be sent after. Every later
+        // round carries the tool calls and their results on top, and none of that is sent again
+        // once the answer stands -- a report of such a round would count a chat far larger than
+        // the one the next request carries. What the answer adds, all rounds of text together, is
+        // counted from its text afterwards, cf. ReportedHistory.
+        //
+        var passesOnUsage = this.internalMessages.Count is 0;
+
         //
         // The text goes out while it is being written; the tool calls are put back together
         // behind it, fragment by fragment.
@@ -62,8 +72,9 @@ public sealed class ChatCompletionToolCallingAdapter<TRequest>(
         await foreach (var serverSentEvent in streamRequestAsync(requestDto, token))
         {
             var part = accumulator.Process(serverSentEvent);
-            if (part.HasContent)
-                yield return ToolCallingStreamEvent.TextDelta(new ContentStreamChunk(part.TextDelta, part.Sources));
+            var usage = passesOnUsage ? part.Usage : TokenUsage.UNKNOWN;
+            if (part.HasContent || usage.IsKnown)
+                yield return ToolCallingStreamEvent.TextDelta(new ContentStreamChunk(part.TextDelta, part.Sources, Usage: usage));
         }
 
         var message = accumulator.Build();

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionToolStreamLine.cs

@@ -11,4 +11,15 @@ namespace AIStudio.Provider.OpenAI;
 /// </remarks>
 /// <param name="Id">The ID of the answer.</param>
 /// <param name="Choices">The choices this line adds to.</param>
-public sealed record ChatCompletionToolStreamLine(string? Id, IList<ChatCompletionToolStreamChoice?>? Choices);
+public sealed record ChatCompletionToolStreamLine(string? Id, IList<ChatCompletionToolStreamChoice?>? Choices)
+{
+    /// <summary>
+    /// What the provider says the request cost, on the one line which carries it.
+    /// </summary>
+    /// <remarks>
+    /// The same block the plain text path reads, and asked for the same way: every streamed
+    /// ChatCompletionAPIRequest asks for it, the requests of the tool rounds included. Not a
+    /// positional parameter, because nobody but the serializer ever builds this line.
+    /// </remarks>
+    public ChatCompletionUsage? Usage { get; init; }
+}