Question 1

Isn't this just a chatbot or an MCP server?

Accepted Answer

No. Chatbots answer questions; MCP is a wire protocol with no permissions or consent layer. Duct is the layer agents act through — with approval steps and an audit trail on every call. Start with the product overview.

Question 2

Why can't I build this directly into my product?

Accepted Answer

You can. Most teams eventually need permissions, approval flows, audit logs, versioning, and revocation. Duct provides those controls as a shared layer instead of rebuilding them for every integration — see action permissions and the threat model.

Question 3

Why not just give the agent an API key?

Accepted Answer

API keys grant broad access. Duct scopes access to specific actions, supports approval flows, and records every invocation.

Question 4

How is this different from just giving my agent an API key?

Accepted Answer

API keys grant broad access with no scoping, no consent step, and no audit trail. Duct issues per-action tokens — an agent approved for get_orders cannot call cancel_subscription, and every call is logged. See agent access.

Question 5

How does Duct compare to MCP or agent frameworks?

Accepted Answer

MCP and function calling define a wire format — you still build auth, permissions, consent, and UI. Agent frameworks build the agent, not what it calls into. Duct lives on the product side: typed manifest, scoped tokens, consent gates, and an embeddable shell.

Question 6

Does Duct replace my existing API?

Accepted Answer

No. Duct sits in front of your API. You map endpoints to actions in the manifest; Duct handles auth, routing, and surfaces. Your backend stays unchanged. See invoke if you are calling from agents.

Question 7

Do we have to rebuild our backend?

Accepted Answer

No. Duct connects to the API you already have. You choose which actions to expose. Execution stays on your servers. The quickstart walks through connecting without a rewrite.

Question 8

Do I have to write the entire manifest by hand?

Accepted Answer

No. duct init generates an initial draft from your project — actions, tiers, and wiring suggestions you review before anything goes live. We also ship an AI integration prompt you can paste into your coding assistant to draft or audit a full Duct setup. The draft is a starting point; the review is the product decision.

Question 9

Do I need to replace my existing auth system?

Accepted Answer

No. Your existing login and session flow stay as-is. Duct issues scoped shell tokens for the embedded widget and agent invoke API — your backend keeps validating requests the way it already does. Auth fields live under configuration → auth.

Question 10

What happens when my API changes after I've set up the manifest?

Accepted Answer

Run duct scan to detect drift between your live API and the manifest. You can push a new manifest version with duct push or restore a previous one with duct pull. Agents always invoke against the live version — see manifest versions.

Question 11

How long until we're live?

Accepted Answer

Most teams are up in under a week. Follow the quickstart, push your manifest, and we provision your shell when your workspace is ready.

Question 12

Can I use Duct in production?

Accepted Answer

Yes. Consent gates, audit logging, and scoped credentials are included from day one — that is what the platform is built for. Review the threat model checklist for how we think about trust boundaries.

Question 13

What happens when an agent needs approval?

Accepted Answer

Sensitive actions can require approval before execution. Duct records who approved, what ran, and the result. Configure this with sideEffects and sensitivity, and see the confirmation protocol for agent callers.

Question 14

How do agents confirm side-effect actions without a UI?

Accepted Answer

The invoke API responds with a structured confirmation step and a short-lived token; the agent retries once approved. Same permission tier and audit trail as the human confirmation card. Details in the confirmation protocol.

Question 15

Is agent identity the same as user consent?

Accepted Answer

No. An agent identity token answers who the caller is. User-scoped work still needs a separate consent token from the user. Destructive work needs an explicit confirmation at call time — prior consent is not enough. See third-party agents and user identity and action permissions by caller.

Question 16

Do chat and the agent API share the same rules?

Accepted Answer

Yes. The same manifest and policy apply whether the request starts in an embedded shell or arrives from a third-party agent via message / invoke. Enforcement differs by caller — see by caller.

Question 17

What shows up in the audit trail?

Accepted Answer

Each attempt leaves a structural record from the enforcement layer: caller identity, action and tier, whether the required proof existed, and the outcome — including consent and confirmation details when those gates applied. Covered under analytics and audit visibility and the audit glossary.

Question 18

Can I revoke access without redeploying?

Accepted Answer

Yes. Revoke a shell key or disable an action in the manifest instantly — no deploy required. In-flight calls using a revoked token are rejected before they reach your API. Cross-shell delegation also supports cascade revocation.

Question 19

What is shell-to-shell?

Accepted Answer

Opt-in cross-product execution. Public shells are discoverable by capability search. Duct mints a single-use delegation token, and both shells log the call with a shared chain ID. Read gate checks, the consent matrix, or the product page.

Question 20

What is a shell?

Accepted Answer

A shell is your product’s Duct surface — the declared actions, permissions, and embeddable chat/API entry points for humans and agents. One product can have one or more shells across environments. See product overview and configuration.

Question 21

Where do I get help if something breaks?

Accepted Answer

Start with the docs and status page. For account or security issues, use the support center — we are a small team and we respond fast. Security reports go through our security overview.

Frequently asked questions

Product

Integration & setup

Permissions, consent & audit

Platform & network

Want Duct on your product before general availability?