AgentInfer 0.3.0

AgentInfer

Object-oriented agents for .NET. An agent is an interface, its prompt is an attribute, and a Roslyn generator writes the implementation at build time.

[Agent("""
    You answer questions about a household ledger.
    Use only figures you were given. Never invent an amount.
    """)]
public interface ILedgerAnalyst
{
    [Prompt("Summarize this spending against its budget in two sentences.")]
    public Task<string> SummarizeAsync(string spending, CancellationToken ct = default);

    [Prompt("Judge whether this summary is supported by the figures given.")]
    public Task<Verdict> ReviewAsync(string summary, string figures, CancellationToken ct = default);
}

That is the whole authoring surface. LedgerAnalystAgent is generated, implements the interface, and takes an AgentRunner. Consumers inject ILedgerAnalyst, so mocking an agent in a test needs no framework support and no model.

Install

dotnet add package AgentInfer

One reference is the whole install: the generator ships inside the runtime package under analyzers/dotnet/cs. Targets net8.0 and net10.0.

The two decisions worth knowing

Predict is the absence of an attribute. [Strategy(Strategies.CodeAct)] is a line somebody typed, that a reviewer sees, and that grep finds. A framework that defaults to executing model-written code runs unreviewed code by default. Opting into code execution has to be visible.

Prompts are string constants, not doc comments. C# strips XML docs into a separate file unreachable at run time. That reads like a handicap and is the opposite: a constant survives compilation and trimming, and cannot fall through to a base class's prompt the way a docstring silently can.

Where it is

The generator, the attributes, a Predict runtime over Microsoft.Extensions.AI, [AgentTool] with compile-time schemas, enforced [RequiresPermission], a tool-calling loop, AgentScope for counting and bounding a whole workflow, and IReplyContract for a typed reply that binds without reflection — all verified end to end against a stub endpoint.

There is no sandbox and no CodeAct. This library chooses which declared tool to call and with what arguments, and nothing more. Treat every [AgentTool] you declare as reachable by anyone who can influence the model's input.

The trim and AOT analyzers are on for everything that ships, so the claim that nothing is discovered at run time is checked by the build rather than asserted here.

Documentation

Build

Needs the .NET 10 SDK — global.json pins it. The library targets net8.0 as well, so CI installs the net8.0 runtime to run that half of the tests.

dotnet build
dotnet test
dotnet run --project samples/Incident     # four workflow patterns, no model needed
dotnet run --project samples/Intake       # a hand-written agent over generated tools
dotnet run --project samples/Ledger       # tools, permissions, model roles

samples/Incident and samples/Intake run against a scripted IChatClient, so they work with nothing installed — the runtime takes an IChatClient and nothing else, which makes a workflow's shape testable without paying for a token. It is not a substitute for a real run: every interesting failure described in the docs came from pointing this at a small local model, and a scripted one reproduces none of them. Pass --live for that, and --metrics to print the instruments at exit.

samples/Ledger needs a real endpoint. See Models and roles for the configuration it reads, and Bounding and measuring for what the counters are for.

No credential belongs in a committed file. Keys come from the environment or from appsettings.Development.json, which is gitignored.

Contributing

Issues and pull requests are welcome. CONTRIBUTING.md covers the part that is not guessable from the tree: the SDK the build pins, the four things that fail in CI more often than anywhere else, when <Version> has to move, and why the generator targets the compiler versions it does.

To report a vulnerability, see SECURITY.md — privately, please, rather than in an issue.

License

MIT. See LICENSE.