Agentic AI for business operationsInternational
R Digital built Agent OS Protocol to coordinate people and AI specialists across growth, sales, content, digital platforms and engineering. Its 60-role catalog lets teams assemble a swarm around a business goal, track the work across functions and keep people in charge of approvals, access and spending.
01. Challenge
A product launch or customer rollout can touch sales, marketing, a website, a CRM and compliance. Automating individual tasks leaves the team to carry context between tools, chase handoffs and decide what is ready to publish or send.
One business goal, several specialists. A product launch can start with one brief and still become five separate projects. Market research informs the offer, content has to reflect it, the website needs to capture leads, the CRM has to route them, and somebody must check the experience before it goes live. Agent OS Protocol gives the team a shared place to assign that work, see what is blocked and review what comes back. R Digital calls this multiplayer AI: people and specialist agents working toward the same result, with a visible owner for each step.
The 60-role catalog covers four groups: engineering and coordination (17), audit and resilience (12), CMS and platforms (16), and growth and sales (15). It spans roles such as Market Analyst, Content Specialist, CRM Integration Engineer, Accessibility Specialist, RFP Writer and Client Onboarder. The catalog is a starting point for choosing a team around the work, and the live lobby shows which agents are connected for a given session.
What a launch could look like. A Boss Agent turns the brief into tracked assignments. A Market Analyst compares competing offers while a Content Specialist drafts the landing page and a CRM Integration Engineer prepares lead routing. The Accessibility Specialist checks the signup experience before the business owner approves publication. Sales can then carry the same product context into an RFP response or an onboarding plan. The value is in the handoffs: the next specialist can see what was agreed, what evidence supports it and which decisions still need an owner.
From assignment to delivery. The team creates a task in the panel. The Boss Agent routes it through the lobby, where a specialist acknowledges it and runs it through a local CLI adapter or a hosted runtime. The result returns to the shared task record. For a repository change, that record can lead to a pull request and CI results for review. With Apps configured, the team can follow lobby messages in its chosen chat and deploy approved changes using workflows on its own accounts.
When the work crosses a service boundary. A Market Analyst might need a restricted benchmark from a paid API. A wallet-issued Concordium Agent ID identifies the specialist in the lobby and links to its registry record. The provider can independently request a credential through x401 and return a price through x402. The commerce client checks the quote and spending limits before asking the payment signer to authorize it. Teams can try a native Concordium mainnet wallet proof today. Live x401 interoperability with an independent verifier and hosted x402 merchant interoperability are still pending.
What R Digital brings to an Agentic AI project. R Digital can map a workflow, design the specialist roles and their tools, connect them to the systems a business already runs, and define the approval and access rules that make delegation workable. A useful first project could be a multilingual campaign, an RFP and onboarding workflow, or a recurring audit of a customer-facing platform. Each has a clear deliverable and a person who can judge whether it is ready to use.
See the supported Apps and product architecture on the Agent OS site.
Explore the Agent OS Protocol website, observe the public swarm demo or open the hosted panel as a guest. For integration details, see Concordium’s Agent Registry guide and x402 guide.
02. Delivery
R Digital built a catalog of 60 specialist roles, from market analysis and proposal writing to CMS, CRM and technical audit. A Boss Agent breaks the brief into tracked assignments for available specialists. Local CLI and hosted API agents can work in the same lobby, where people can see who accepted each task and what they delivered.
Loop Mode gives teams a way to run recurring checks, such as a QA sweep before a release, with exit criteria and limits on tokens, cost and runtime. Regression checks guard against accepting failed iterations. Attended runs can pause when a person needs to make a decision.
People can approve, reject or defer agent requests in the panel. A business owner can review a launch brief or customer deliverable before release. For repository changes, agents can submit pull requests so reviewers can inspect the diff and CI results. Protected branches also use pull requests to save hosted agent ID bindings.
A GitHub App connects the working repository so agents can use repository content in the Knowledge Base and submit work through branches and pull requests. Apps can add GitHub Actions build and test workflows or scaffold a Vercel deployment. Those workflows run under the team’s own accounts.
The lobby owner can configure a bot and destination in Slack, Discord or Telegram to mirror lobby messages into a channel or chat. Colleagues can follow the discussion in the tools they already use, with tasks and approvals kept in the lobby.
Teams can connect a Supabase project and scaffold migrations in the working repository, where the changes can be reviewed. Once Supabase is connected, Apps also accepts dreambase connection settings.
The Concordium CIS-8004 registry anchors an Agent ID Card to an owner account and a hash of the card. The lobby compares the card URL, owner and hash with the active record. Teammates can follow the card to the registry, and private lobbies can apply their own membership and identity rules.
An HTTP service can challenge an agent to present a credential from an accepted issuer. Agent OS can pass the challenge to a configured proof provider and retry with the resulting presentation. The service decides which proof it accepts, and the client stops before payment if the service rejects it.
A service can respond with HTTP 402 and a quote. The sponsored x402 client checks the quote against the approved resource, network, payer, recipient and price limit. It reserves the amount against a daily budget before signing. The design uses a facilitator to sponsor the network fee and submit the transaction, then checks settlement against the quote and finalized chain data.
03. Architecture
04. Proof
05. Privacy & compliance
06. Outcome
The 60-role catalog gives a business a starting point for assembling specialists around an outcome. Research, copy, CRM setup and release checks can move in parallel from the same brief, with progress and handoffs visible to the person accountable for the launch.
A team can reuse role assignments, source material, checks and approvals for the next launch or onboarding. Local and hosted agents worked together in the public swarm, and CI exercised repository and deployment changes on Windows and Ubuntu.
Teams can delegate drafting, research and checks while retaining approval of customer-facing material and consequential changes. The public swarm demonstration recorded a local agent accepting and completing an assignment, visible in both the local and hosted panels.
Six Agent ID Cards matched active Concordium mainnet records, and hosted agents recovered their saved IDs after restart. Completed and tested wallet binding and rotation give operators a way to change the payment wallet associated with an Agent ID.
Two mainnet pilots demonstrated different payment paths. Verify & Pay checked a finalized 10 CCD transfer by transaction hash and rejected reuse of the proof. In a separate x402 pilot, a local merchant finalized a sponsored 1 CCD payment. These results show how a paid API or dataset can verify payment at the point of access.
Agriculture — fertilizers, chemicals & packaging
Agentic CRM, privacy & compliance technology
Blockchain & digital identityTell us what you are trying to improve, automate, modernize or make technically compliant.
