TypeshiftAgents and tools, defined in code

Build AI agents for .NET with just a few lines of code, and let the compiler catch your mistakes.

Design your toolsets

Group your tools as you like, by feature, domain or permission level. A tool is a static method you add with one line, and its XML docs tell the model what it does.

Write a toolset

Create an agent

A few chained calls in the agent's constructor pick the model, the instructions, the tools and their settings.

Write an agent

Use an agent in your app

Inject the agent's host into a controller or an endpoint, or resolve it in a background job, and its tools use that request's or job's services.

Use an agent

Catch mistakes at compile time

Configure a tool the agent does not offer, and the build fails. The analyzer reports the rest, each with how to fix it.

See the diagnostics

Store settings per customer

Each agent gets a typed Settings record of what you mark configurable. Store it in your own database, under your own ids.

Store settings

Keep conversations that survive a crash

Every step is stored as it happens, so a run resumes where it stopped. Steer a run while it works, or watch it live.

Keep a conversation

Approve, plan and delegate

Tool approval, plan mode, sub-agents, compaction and MCP servers.

Approve tool calls

Publish with Native AOT

The generator writes every path a run takes, so nothing reflects, and the analyzer reports a tool type that lacks JSON metadata.

Publish with AOT