SharpCoder.Providers 0.22.1

SharpCoder.Providers

SharpCoder.Providers supplies the IChatClient factory used by SharpCoder. It creates clients for GitHub Copilot, GitHub Models, and Ollama. Keep the SharpCoder and SharpCoder.Providers package versions aligned when using them together.

Install

dotnet add package SharpCoder.Providers

Create a client

Pass a provider-prefixed model string to select a provider explicitly. The factory returns Microsoft.Extensions.AI.IChatClient:

using SharpCoder.Providers;

using var client = ChatClientFactory.Create("copilot/<model>");

Replace <model> with the model identifier you want to use. Without a recognized prefix, LLM_PROVIDER selects the provider (default: copilot), and the remainder, if supplied, is treated as the model identifier.

Providers and configuration

Supported prefixes are copilot, github, ollama-cloud, and ollama-local. A provider prefix is case-insensitive; only these recognized prefixes before / select a provider.

For Copilot, ChatClientFactory.SetTokenProvider(Func<string?>) takes precedence when it returns a non-whitespace token; otherwise the factory checks GH_TOKEN, then GITHUB_TOKEN. Whitespace values are treated as absent. ChatClientFactory.IsTokenAvailable() reports whether a non-whitespace Copilot token is available using that same precedence. The GitHub Models provider uses GH_TOKEN and GITHUB_TOKEN only; it does not use the token provider.

Copilot endpoint discovery

For Copilot requests, the factory can discover the account's advertised API endpoint. Discovered endpoints are accepted only for HTTPS hosts at githubcopilot.com or its subdomains; other values fall back to ChatClientFactory.DefaultCopilotApiEndpoint (https://api.githubcopilot.com). Discovery failures also fall back to that default. The public ChatClientFactory.GetCopilotApiEndpointAsync(string token, CancellationToken) exposes the lookup. GitHub Enterprise (*.ghe.com) is not supported.

Built-in behavior

• Copilot and both Ollama providers use a resilience pipeline with up to three exponential-backoff retries and a 20-minute pipeline timeout; GitHub Models does not use this pipeline.
• Reasoning effort is mapped at the provider boundary: extra_high maps to xhigh for Copilot and max for Ollama; GitHub Models clamps it to high.
• Copilot request/response diagnostics are opt-in. Use ChatClientFactory.SetDiagnosticsDirectory(path) or set SHARPCODER_DIAGNOSTICS_DIR; an explicit method setting takes precedence. These diagnostic files are not produced by GitHub Models or either Ollama provider.

License and source

Licensed under the MIT License.

SharpCoder repository