Deliverd

Introduction

What Deliverd does, who it is for, and where to start: the guide for the people who answer agents and the administrators who run Deliverd for an organisation.

Deliverd is human-in-the-loop infrastructure for AI agents: where your organisation's AI agents meet the people they work for. An agent that needs a person — for an approval, a decision, an answer or a review — asks, a person answers on a page that works on a phone, and the agent carries on with the answer. Who decided what, and when, is kept. When an agent has made something people need to read, it can also publish it to a secure address behind your organisation's own sign-in.

What an agent can ask for

Approvals, reviews and requests
An approval before a consequential action, a review of a draft, or typed questions to the people who know the answer. The person decides on a page, from email, in Slack or Teams, or from a notification on their phone, and the agent receives the answer. See Deciding an approval and Reviews and requests for information.
Gates
Before a registered action, an agent asks whether it may go ahead at all. Your organisation's policy answers allow, deny, or a person decides — and nothing is allowed because nobody thought about it. See Policies.
Objectives
A job's objective and state, kept apart from the conversation doing the work, so an agent that resumes in another tool picks up where it left off. See Objectives and side tasks.
Publishing
What an agent made — a report, a dashboard, a briefing — on an address inside your organisation's sign-in, with every version kept, comments that reach the agent, and a PDF when someone needs a file. See Publishing and sharing reports.
Evidence
Every request leaves a record: who was asked, what they were told, what they said and when. A project groups one job's requests, tasks and gates into a single evidence timeline — the API calls it a flow — and an evidence pack is that timeline as a file. See Projects and evidence packs.

Who this guide is for

You answer agentsDeciding approvals, reviewing drafts, answering requests, reading reports.You run DeliverdPeople, groups, sign-in, policies, ethics rules, integrations and billing.You are setting it upThe order to do things in, from inviting people to connecting the first agent.You connect agentsMCP, the REST API, the TypeScript and Python SDKs, and the CLI live in the developer docs.

How a request travels

  1. 1

    An agent asks

    Over MCP, the REST API, the TypeScript and Python SDKs, or the CLI. It says what it wants to do, why, and who should decide.

  2. 2

    Your organisation's rules read it

    Policies decide who must agree and how many. Ethics rules can put a concern in front of the approver. None of them can approve anything on their own.

  3. 3

    A person decides

    In their inbox, by email, in Slack or Teams, or after a notification on their phone. They can approve, reject, correct the details, or ask a question first.

  4. 4

    The agent carries on

    It receives the decision, and anything the approver changed, and acts on it. Who decided, when and why is kept in the audit log.

Try it on your own agent.

Free to start, no card. Sign up with Google, Microsoft or GitHub, and the first page after hands you an API key.

Start free →