FakeModelProvider

Namespace Tracon.Testing · Assembly Tracon.Testing.dll

Configurable, non-networked fake model provider.

public sealed class FakeModelProvider : ITenantCredentialModelProvider, IModelProvider, IDisposable

Inheritance

object ← FakeModelProvider

Implements

ITenantCredentialModelProvider, IModelProvider, IDisposable

Inherited Members

object.GetType(), object.ToString(), object.Equals(object?), object.Equals(object?, object?), object.ReferenceEquals(object?, object?), object.GetHashCode()

Examples

var provider = new FakeModelProvider().
CallsTool("get_order_status", new { orderId = "ORD-7" }).
EchoesUserMessage();

Remarks

Five fluent methods (FakeModelProvider.RespondsWith, FakeModelProvider.EchoesUserMessage, FakeModelProvider.CallsTool, FakeModelProvider.ForModel, FakeModelProvider.WithModel) cover all the fake provider behavior that is today duplicated across five separate files in Tracon’s test suites.

Each model has its own ordered response queue (selected with FakeModelProvider.ForModel; the default queue is used when none is selected). A call pops the next step in the queue; once the queue is drained, every subsequent call returns either the echo set with FakeModelProvider.EchoesUserMessage or a fixed response. This behavior is lasting for the provider’s lifetime — it does not infer “which tool was already called” by scanning message history.

Constructors

FakeModelProvider(string)

Creates a new fake provider.

public FakeModelProvider(string name = "fake")

Parameters

name string

Provider name.

Properties

Models

The models this provider offers.

public IReadOnlyList<ModelDescriptor> Models { get; }

Property Value

IReadOnlyList<ModelDescriptor>

Remarks

Metadata, not an allow list — see the remarks on IModelProvider. This is read on the compile path and may be read concurrently; return a stable, immutable collection rather than one that is mutated after construction.

Name

The provider name. ModelBinding.Provider matches this value. Comparison is case-insensitive.

public string Name { get; }

Property Value

string

Requests

Requests that reached this provider; the newest is last.

public IReadOnlyList<FakeModelRequest> Requests { get; }

Property Value

IReadOnlyList<FakeModelRequest>

Methods

CallsTool(string, object?)

Enqueues a tool call.

public FakeModelProvider CallsTool(string toolName, object? arguments = null)

Parameters

toolName string

Name of the tool to call.

arguments object?

Call arguments. If not an IDictionary, its properties are read through reflection (works for anonymous types).

Returns

FakeModelProvider

The chain, for continued configuration.

CallsTools(params (string ToolName, object? Arguments)[])

Enqueues a SINGLE turn that calls every given tool at once.

public FakeModelProvider CallsTools(params (string ToolName, object? Arguments)[] calls)

Parameters

calls (string ToolName, object? Arguments)[]

The tool name and arguments for each call.

Returns

FakeModelProvider

The chain, for continued configuration.

Remarks

FakeModelProvider.CallsTool enqueues one call per turn and can never produce more than one FunctionCallContent in the same response; exercising a REAL concurrent tool-call loop (FunctionInvokingChatClient.AllowConcurrentInvocation) needs several independent calls to arrive together, in one turn.

CreateChatClient(ModelBinding)

Produces a raw chat client for the given binding.

public IChatClient CreateChatClient(ModelBinding binding)

Parameters

binding ModelBinding

The model binding.

Returns

IChatClient

The provider-specific client. Decorators specific to the provider (example: Anthropic’s settings decorator) may be added here.

Remarks

Do not build the common pipeline here. UseFunctionInvocation, UseOpenTelemetry, the content guard, the circuit breaker, the per-provider concurrency limiter, attachment resolution, response caching, the fallback chain and content-filter detection are all added by ModelProviderRegistry.CreateChatClient. Before that shared pipeline, every provider package built the tool-call loop inside itself; the result was that no ring the registry wraps around could see the loop’s turns — a tool result entered the model uninspected.

If the loop is also built here, two nested FunctionInvokingChatClient instances form: the inner one resolves tools, the outer one never sees any call. This is not merely cosmetic. The registry places the content guard directly above the client this method returns, so that every turn of the tool-call loop is inspected. With an inner loop, the turn that carries a tool result back into the model runs beneath the guard: measured on a single-tool run, the guard sees the tool result only on its way out (ContentGuardDirection.Output) and never on its way in (ContentGuardDirection.Input) — which is the exact path prompt injection takes. The reply text and the tool-call count are unchanged, so nothing else reveals the mistake.

Lifetime of the returned client. Tracon does not dispose it. The client is built once per compiled agent and stored in it; the compiled agent (Microsoft.Agents.AI.ChatClientAgent) implements neither IDisposable nor IAsyncDisposable, and evicting an agent from the compile cache drops the reference without disposing. An implementation therefore owns the lifetime of whatever it returns, and what it returns must tolerate never being disposed. This is why the shipped providers return clients backed by a long-lived, shared SDK client rather than a per-call one: returning a client that holds a resource needing release would leak it.

CreateChatClient(ModelBinding, ModelProviderCredential)

Produces a raw chat client using a resolved tenant credential.

public IChatClient CreateChatClient(ModelBinding binding, ModelProviderCredential credential)

Parameters

binding ModelBinding

The model binding.

credential ModelProviderCredential

The non-null tenant credential.

Returns

IChatClient

The provider-specific raw chat client.

Remarks

The implementation uses ModelProviderCredential.ApiKey and, when present, ModelProviderCredential.Endpoint. The API key must never fall back to the setup-time key. An endpoint may fall back to a setup-time endpoint when only the key is overridden.

Dispose()

public void Dispose()

EchoesLastToolResult(string, int?, int?)

Starts echoing the LAST tool result in history once the queue is drained.

public FakeModelProvider EchoesLastToolResult(string prefix = "", int? inputTokens = null, int? outputTokens = null)

Parameters

prefix string

Text prepended to the result.

inputTokens int?

If given, the input token count reported with every response.

outputTokens int?

If given, the output token count reported with every response.

Returns

FakeModelProvider

The chain, for continued configuration.

Remarks

For the last step of a tool chain (set up with FakeModelProvider.CallsTool), where the final response must genuinely depend on the result the tool returned — e.g. a hand-off result to a sub-agent.

EchoesUserMessage()

Starts echoing the last incoming user message once the queue is drained.

public FakeModelProvider EchoesUserMessage()

Returns

FakeModelProvider

The chain, for continued configuration.

ForModel(string, Action<FakeModelProvider>)

Defines a separate response queue for a specific model name. Calls to FakeModelProvider.RespondsWith/FakeModelProvider.EchoesUserMessage/FakeModelProvider.CallsTool inside the configure body affect only that model’s queue.

public FakeModelProvider ForModel(string modelId, Action<FakeModelProvider> configure)

Parameters

modelId string

Model name.

configure Action<FakeModelProvider>

Configuration that fills this model’s queue.

Returns

FakeModelProvider

The chain, for continued configuration.

RespondsWith(params string[])

Enqueues responses to return in order. Once the queue is drained, a fixed response is returned unless FakeModelProvider.EchoesUserMessage was set.

public FakeModelProvider RespondsWith(params string[] responses)

Parameters

responses string[]

Texts to return in order.

Returns

FakeModelProvider

The chain, for continued configuration.

RespondsWith(string, int, int)

Enqueues a single text response that reports a specific token usage.

public FakeModelProvider RespondsWith(string response, int inputTokens, int outputTokens)

Parameters

response string

Text to return.

inputTokens int

Input token count to report.

outputTokens int

Output token count to report.

Returns

FakeModelProvider

The chain, for continued configuration.

Remarks

For scenarios that test cost/usage metrics end to end.

WithModel(ModelDescriptor)

Registers a model on this provider.

public FakeModelProvider WithModel(ModelDescriptor descriptor)

Parameters

descriptor ModelDescriptor

Model descriptor.

Returns

FakeModelProvider

The chain, for continued configuration.