MindWorkAI/AI-Studio
1
0 Fork 0
mirror of https://github.com/MindWorkAI/AI-Studio.git synced 2026-10-17 15:18:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add llmman as a self-hosted host for chat and embedding providers (#947)

Browse Source 
Co-authored-by: Thorsten Sommer <SommerEngineering@users.noreply.github.com>
This commit is contained in:
Eric CurtinandThorsten Sommer authored and GitHub committed 2026-10-13 13:52:43 +02:00
1 parent 2bd3d41421
commit 88d1f4d96a
20 files changed
+270 -89
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

app/MindWork AI Studio/Provider/SelfHosted/ProviderSelfHosted.cs
+40 -55
View File
@@ -25,7 +25,7 @@ public sealed class ProviderSelfHosted(Host host, string hostname) : BaseProvide
public override string InstanceName { get; set; } = "Self-hosted";
/// <inheritdoc />
public override bool HasModelLoadingCapability => host is Host.OLLAMA or Host.LM_STUDIO or Host.VLLM or Host.LLAMA_CPP;
public override bool HasModelLoadingCapability => host.IsModelListSupported();
/// <inheritdoc />
public override async IAsyncEnumerable<ContentStreamChunk> StreamChatCompletion(Provider.Model chatModel, ChatThread chatThread, SettingsManager settingsManager, [EnumeratorCancellation] CancellationToken token = default)
@@ -40,7 +40,7 @@ public sealed class ProviderSelfHosted(Host host, string hostname) : BaseProvide
{
// Build the list of messages. The image format depends on the host:
// - Ollama uses the direct image URL format: { "type": "image_url", "image_url": "data:..." }
// - LM Studio, vLLM, and llama.cpp use the nested image URL format: { "type": "image_url", "image_url": { "url": "data:..." } }
// - LM Studio, vLLM, llama.cpp, and llmman use the nested image URL format: { "type": "image_url", "image_url": { "url": "data:..." } }
var messages = host switch
{
Host.OLLAMA => await chatThread.Blocks.BuildMessagesUsingDirectImageUrlAsync(this.CreateSettingsProvider(effectiveChatModel)),
@@ -94,22 +94,17 @@ public sealed class ProviderSelfHosted(Host host, string hostname) : BaseProvide
{
try
{
switch (host)
if (host is Host.LLAMA_CPP)
return await this.LoadLlamaCppTextModels(apiKeyProvisional, token);
if (!host.IsChatSupported() || !host.IsModelListSupported())
return ModelLoadResult.FromModels([]);
var result = await this.LoadModels(SecretStoreType.LLM_PROVIDER, apiKeyProvisional, token);
return result with
{
case Host.LLAMA_CPP:
return await this.LoadLlamaCppTextModels(apiKeyProvisional, token);
case Host.LM_STUDIO:
case Host.OLLAMA:
case Host.VLLM:
var result = await this.LoadModels(SecretStoreType.LLM_PROVIDER, apiKeyProvisional, token);
return result with
{
Models = [..result.Models.Where(model => model.IsChatModel(this.Provider))]
};
}
return ModelLoadResult.FromModels([]);
Models = [..result.Models.Where(model => model.IsChatModel(this.Provider))]
};
}
catch(Exception e)
{
@@ -128,19 +123,14 @@ public sealed class ProviderSelfHosted(Host host, string hostname) : BaseProvide
{
try
{
switch (host)
{
case Host.LM_STUDIO:
case Host.OLLAMA:
case Host.VLLM:
var result = await this.LoadModels(SecretStoreType.EMBEDDING_PROVIDER, apiKeyProvisional, token);
return result with
{
Models = [..result.Models.Where(model => model.IsEmbeddingModel(this.Provider))]
};
}
if (!host.IsEmbeddingSupported() || !host.IsModelListSupported())
return ModelLoadResult.FromModels([]);
return ModelLoadResult.FromModels([]);
var result = await this.LoadModels(SecretStoreType.EMBEDDING_PROVIDER, apiKeyProvisional, token);
return result with
{
Models = [..result.Models.Where(model => model.IsEmbeddingModel(this.Provider))]
};
}
catch(Exception e)
{
@@ -154,33 +144,28 @@ public sealed class ProviderSelfHosted(Host host, string hostname) : BaseProvide
{
try
{
switch (host)
{
case Host.WHISPER_CPP:
return ModelLoadResult.FromModels(
[
new Provider.Model("loaded-model", TB("Model as configured by whisper.cpp")),
]);
//
// These two answer the models endpoint with everything they serve, and nothing in
// that answer says which of them listens. Asking what each model is made for is the
// only thing standing between this list and every chat and embedding model of the
// installation, which is what it used to hold. An engine running no speech model at
// all therefore offers nothing here, and says so, rather than offering models which
// would fail the moment audio reaches them.
//
case Host.OLLAMA:
case Host.VLLM:
var result = await this.LoadModels(SecretStoreType.TRANSCRIPTION_PROVIDER, apiKeyProvisional, token);
return result with
{
Models = [..result.Models.Where(model => model.IsTranscriptionModel(this.Provider))]
};
if (host is Host.WHISPER_CPP)
return ModelLoadResult.FromModels(
[
new Provider.Model("loaded-model", TB("Model as configured by whisper.cpp")),
]);
default:
return ModelLoadResult.FromModels([]);
}
if (!host.IsTranscriptionSupported() || !host.IsModelListSupported())
return ModelLoadResult.FromModels([]);
//
// The engines which get here answer the models endpoint with everything they serve, and
// nothing in that answer says which of them listens. Asking what each model is made for
// is the only thing standing between this list and every chat and embedding model of
// the installation, which is what it used to hold. An engine running no speech model at
// all therefore offers nothing here, and says so, rather than offering models which
// would fail the moment audio reaches them.
//
var result = await this.LoadModels(SecretStoreType.TRANSCRIPTION_PROVIDER, apiKeyProvisional, token);
return result with
{
Models = [..result.Models.Where(model => model.IsTranscriptionModel(this.Provider))]
};
}
catch (Exception e)
{