Deliverd

Changelog

What shipped.

Everything here is live in production today. What is built but not yet available to you, such as a store listing still in review, is left out until it is.

    • Plans carried through sign-up

      Choosing a paid plan on the pricing page now takes you through sign-up and straight to that plan's checkout, with its 14-day trial.

      Read more

    • AI register

      Each AI system has a lifecycle, a risk classification, the controls it owes, decisions from your review forum and review dates, with reports in your own branding.

      Read more

    • Projects and signed evidence packs

      Tie approvals, reviews, questions and publishes to a project and export it as one evidence pack. The audit trail is hash-chained per organisation, and packs are signed with a published key, so anyone can check a copy offline.

      Read more

    • Business Starter plan

      Policy rules for agents, including two-approver rules by risk, and sign-off before a report leaves, for up to 20 seats.

      Read more

    • Deliverd Screen, open source

      The detectors behind data screening, as an MIT-licensed library for TypeScript, Python and Rust: find and redact personal data, secrets and your own confidential terms before text goes to a model, on your own machine.

      Read more

    • Data screening on Free, report-only

      Free organisations see what data screening would have held back, without anything being held.

      Read more

    • GitHub Actions

      Pause a workflow until a person approves, with the decision on the record.

    • Muse Spark, Mastra, CrewAI and Pydantic AI

      Four more agent frameworks and platforms on the integrations page, each with a setup guide.

    • Supplier AI due diligence, standard questions

      A standard question set for assessing a supplier's AI, available to every Compliance Mode organisation, with exports in your own branding.

    • AI activity and control report

      Analytics show how concentrated your AI use is across vendors, and download as a branded report.

    • Compliance Mode in the SDKs

      The organisation-wide Compliance Mode API is now in the TypeScript and Python SDKs.

      Read more

    • Data screening before a model

      Check a prompt for personal data, secrets and confidential names against what your AI register approves that model for, then allow, redact, ask a person or refuse. Detection runs where your agent runs, so the text never has to reach us.

      Read more

    • Supplier AI due diligence

      Classify a supplier's AI, send them questions, record an EU AI Act determination, sign off by tier and set the next review date.

    • Approve from the lock screen

      Approve or reject from a phone notification, and require a passkey for high-risk decisions.

      Read more

    • Amazon Bedrock, Google Vertex AI and Muse

      Agents on Amazon Bedrock and Google Vertex AI can ask people through Deliverd, and Muse Code and Muse Spark are supported in setup and the SDKs.

    • Security fixes from an internal review

      Fixes from an internal adversarial review of the product. The gaps still open are listed on the security page.

      Read more

    • Shopify, Wix and Hoplite

      Online stores and Hoplite agents can route their questions to people.

    • Ask Deliverd

      A chat assistant on the website that explains the product and hands over to a person.

    • Rust SDK

      The deliverd crate on crates.io, alongside the TypeScript and Python SDKs.

      Read more

    • Integrations page

      Every integration in one catalogue, with official logos, the same in the product and on the website.

    • SDKs published from a public repository

      The SDKs and the CLI are built and published from a public repository, with the source for every release open to read.