For Developers · AgentArc

Typed SDK

First-class TypeScript and Python. Tools, schemas, and state are typed end to end — autocomplete the whole way.

Local dev loop

Run agents on your machine with mocked tools and recorded fixtures. Iterate fast before anything hits the cloud.

Evals in CI

Assert on agent behavior with eval suites that run on every PR — catch regressions before they ship.

Versioned deploys

Every deploy is immutable and tagged. Promote, roll back, and run blue/green between agent versions.

Config as code

Define agents, tools, and guardrails in version control. Review changes in PRs like any other infra.

Open standards

MCP for tools, OpenTelemetry for traces, OpenAI-compatible model APIs. No lock-in to fight later.

Quickstart

From zero to a running agent

Install the CLI, define an agent, run it locally, then deploy. The same definition runs everywhere.

One CLI for dev, test, and deploy.
Bring your own model keys — no inference markup.
Self-host the runtime in your own VPC anytime.

~/my-agent

# install & scaffold
$ npm i -g @agentarc/cli
$ agentarc init triage-bot

# run locally with mocked tools
$ agentarc dev

# ship it
$ agentarc deploy --prod
✓ deployed v1 · https://run.agentarc.dev

Build agents like
you build software.