Start with the outcome
Name the agent, its business area and the outcome you are aiming for. Required guardrails are added automatically.
AI Foundry
AI Foundry is the design environment for AI agents. Build agents and co-pilot apps step by step, with required guardrails locked in from the first step and a governed path from prototype to production.
Foundry keeps the builder simple and the guardrails non-negotiable.
Name the agent, its business area and the outcome you are aiming for. Required guardrails are added automatically.
Add a trigger, AI reasoning, connectors, decisions and human review, the same way you would write a recipe.
Foundry connects to AI Harness while you build, so the guardrails, tier and checks come from the governance layer, not from guesswork.
Move from prototype to pilot to production through named gates, each with a recorded human sign-off.
Two kinds of builds, one set of guardrails.
A backend flow that runs without a person typing: a trigger, AI steps, connectors, decisions and a point where a person signs off.
A chat assistant people talk to directly. Describe it in plain language and watch a live preview take shape: name, greeting, model and sample prompts.
Trigger, AI reasoning, guardrail, connector, decision and human review. Steps can be reordered at any time.
PII redaction is built into every agent. Co-pilot apps also get prompt-injection screening and an audit log, locked in from the start.
A human review step defines who reviews and what they can do: approve, edit or reject. The agent cannot finalize past that point.
Choose read-only, write to one field, or write to a full record. Full write access needs security and compliance sign-off before it can go live.
Promotion runs readiness checks and writes every step to a ledger: who, what and when.
Pick the model for each step, add an optional fallback, and see the status of every model and platform before you commit.
Examples of the kinds of agents and co-pilot apps teams design in Foundry.
Foundry designs. Harness governs. Compliance decisions, risk tiers and approvals come from Harness, and Foundry shows them as you build.
Explore AI HarnessWe are building Foundry with design partners. These are on the way.
Start from reviewed patterns instead of a blank page.
Generate the agent specification and playbook from the design.
Run test sets and keep the results as release evidence.
Take an agent definition to another supported cloud.
Foundry is designed to work with leading models and with AWS, Azure and Google Cloud.
Tell us what you want an agent to do, and we will show you how it takes shape in Foundry.