# Delegrity

> Delegrity checks, at the moment an AI agent acts, that your organisation still authorises that exact action, and keeps signed evidence of what happened.

Delegrity is the authority layer between AI agents and business systems. Identity, token and connection checks establish who an agent is; Delegrity checks whether this exact action is still authorised, just before it runs.

## How it works

- Bind: the agent's request becomes one exact, fingerprinted action.
- Decide: the mandate is checked, the customer's own policy engine is consulted and an approver is asked where required.
- Grant: the action is rechecked just before execution; single-use authority is issued only if nothing material changed.
- Prove: a signed, hash-chained receipt records what was decided and what happened.

Decision values in the contract: AUTHORIZED, AWAITING_APPROVAL, BLOCKED, EXPIRED.

## Status

- Delegrity is at design-partner stage.
- No production customer evidence and no independent certification yet.
- No named vendor is connected, and none is implied.
- Coordination of authority between multiple agents is studied, not built.
- Customers can define their own actions, held for a person by default; running them against a real system is still in test.

## Pages

- [Home](https://delegrity.com/): the problem, how it works, the developer endpoint, the security review boundary and an FAQ.
- [Blog](https://delegrity.com/blog): articles on authority, approval and evidence for AI agents.
- [Functions](https://delegrity.com/functions): what Delegrity does, one page per function.
  - [Mandates](https://delegrity.com/functions/mandates): Written authority: who the agent acts for, on what, until when, and up to how much.
  - [Exact action](https://delegrity.com/functions/exact-action): Every request becomes one fingerprinted action, so nothing can quietly change later.
  - [Human approval](https://delegrity.com/functions/human-approval): A person decides on the exact values, and the agent can never approve itself.
  - [Execution checkpoint](https://delegrity.com/functions/execution-checkpoint): Just before anything runs, authority is checked again, and a single-use grant is issued.
  - [Revocation](https://delegrity.com/functions/revocation): Take authority away and the agent stops, even with a working login and token.
  - [Outcome check](https://delegrity.com/functions/outcome-check): Approved is not done. Delegrity reads back what the business system actually did.
  - [Signed receipts](https://delegrity.com/functions/signed-receipts): A signed, hash-chained record of who decided what, that can be checked independently.
  - [Emergency controls](https://delegrity.com/functions/emergency-controls): Hold one agent or one action for up to an hour, and keep the record of why.
  - [Connect your agents](https://delegrity.com/functions/connect-agents): One authority layer for the different ways agents reach your business systems.
  - [Observe mode](https://delegrity.com/functions/observe-mode): See what Delegrity would decide on real agent traffic before it controls anything.
  - [Your own actions](https://delegrity.com/functions/your-own-actions): Govern the actions your organisation defines, not only the ones built in.
  - [Connectors](https://delegrity.com/functions/connectors): Connect a business system once, and every action to it goes through the checkpoint.
- [Solutions](https://delegrity.com/solutions): the actions an agent must not get wrong.
  - [Supplier bank changes](https://delegrity.com/solutions/supplier-payments): Run against a real payment provider in test mode.
  - [Refunds](https://delegrity.com/solutions/refunds): Rules tested. No real refund system connected yet.
  - [Access changes](https://delegrity.com/solutions/access-changes): Rules tested. No real identity system connected yet.
  - [Code deploys](https://delegrity.com/solutions/code-deploys): Rules tested. No real deploy pipeline connected yet.
- [Developers](https://delegrity.com/developers): quickstart, concepts, SDKs and API reference.
  - [Quickstart](https://delegrity.com/developers/md/quickstart): Propose one action and read what Delegrity decided.
  - [Core concepts](https://delegrity.com/developers/md/concepts): The few ideas everything else builds on.
  - [Ways in](https://delegrity.com/developers/md/ways-in): The different routes an agent can take to Delegrity.
  - [Set up with a coding agent](https://delegrity.com/developers/md/agent-setup): Paste this into your coding agent to wire in Delegrity.
  - [Wait for a result](https://delegrity.com/developers/md/results): Follow one action from decision to receipt.
  - [Offers from people](https://delegrity.com/developers/md/offers): When a person asks your agent to do something.
  - [Errors and retries](https://delegrity.com/developers/md/errors): What each failure means and whether to retry.
  - [Your own actions](https://delegrity.com/developers/md/your-own-actions): Govern an action your organisation defines, not only the built-in ones.
  - [Observe mode](https://delegrity.com/developers/md/observe-mode): See what Delegrity would decide on real traffic before it controls anything.
  - [Connect a system](https://delegrity.com/developers/md/connect-a-system): What a business system must offer before your own actions can run on it.
  - [Check a request](https://delegrity.com/developers/md/check-a-request): See how Delegrity reads a request before your agent sends it. Nothing is stored or run.
  - [Python SDK](https://delegrity.com/developers/md/python-sdk): Propose actions and read results from Python.
  - [TypeScript SDK](https://delegrity.com/developers/md/typescript-sdk): Propose actions and read results from Node.js.
  - [API reference](https://delegrity.com/developers/md/api-reference): The four endpoints an agent uses.
  - [Glossary](https://delegrity.com/developers/md/glossary): The words these docs use, in one line each.
- [Test results](https://delegrity.com/proof): what we test on every release.
- [How it works](https://delegrity.com/how-it-works): one action, from the request to a signed receipt.
- [How we compare](https://delegrity.com/approaches): beside identity, policy engines, audit logs and agent gateways.
- [For security teams](https://delegrity.com/solutions/for/security-teams): Agents act through access they already have. Delegrity decides whether this exact action may run.
- [For finance operations](https://delegrity.com/solutions/for/finance-operations): An agent that moves money must move exactly what a person approved, once.
- [For platform engineering](https://delegrity.com/solutions/for/platform-engineering): One call before every protected action, from whichever way your agents already work.
- [Trust](https://delegrity.com/trust): what we can show, and what we don't claim.
- [Contact](https://delegrity.com/contact): talk to us.
- [About](https://delegrity.com/about): who we are and where we stand.
- [Design partners](https://delegrity.com/design-partners): how working with us starts.
- [Privacy notice](https://delegrity.com/privacy)

## Articles

- [When two AI agents are both authorised, who decides?](https://delegrity.com/blog/when-authorised-agents-collide): Single-agent controls ask whether an agent may act. Systems of agents add a harder question: what happens when two legitimately authorised agents collide.
- [Access is not authority: why valid tokens still allow wrong actions](https://delegrity.com/blog/access-is-not-authority): An agent can pass every identity and token check and still take an action nobody authorised. Here is why, and what a per-action authority check adds.
- [What the 2026 data says about AI agents acting without authority](https://delegrity.com/blog/agents-acting-without-authority): Surveys, incidents and new guidance point to one gap: agents with valid access taking actions nobody authorised, and organisations unable to show what happened.
- [What changes after approval? The beneficiary-swap problem in AI agents](https://delegrity.com/blog/beneficiary-swap-problem): An approval is only as good as its link to the action that runs. Why approving a request type is not enough, and how a recheck before execution closes the gap.
- [Mandates, not permissions: how organisations should delegate to AI agents](https://delegrity.com/blog/mandates-not-permissions): Permissions say what an agent can reach. A mandate says what it may do, for whom, within which limits, and until when. Why delegation to agents needs the second.

## Contact

- Become a design partner: https://delegrity.com/design-partners
- Email: avinash@delegrity.com