A2A
A2A (Agent2Agent) is an open protocol for agents to discover each other, exchange messages and work on tasks together, whatever platform they run on. In Orqane, A2A works both ways: you publish your employees to outside agents, and your employees send tasks to outside agents.
How A2A works
Section titled “How A2A works”| A2A concept | What it is | In Orqane |
|---|---|---|
| Agent Card | A JSON document, served at /.well-known/agent-card.json, that describes an agent: its name, its skills, its endpoint, how to authenticate and whether it supports streaming |
Generated from the employee’s profile and skills |
| Task | A unit of work with a lifecycle: submitted, working, input-required, completed, failed, canceled, rejected |
Maps to an Orqane task and its runs |
| Message | A turn in the conversation, made of text, file or data parts | Appears in the task’s conversation |
| Artifact | An output the agent produces, such as a report or a file | Attached to the task result |
| Streaming | Live updates on a task as it progresses | Sent as the employee completes steps |
Publish your employees
Section titled “Publish your employees”Turn it on
Section titled “Turn it on”Enable inbound A2A
An Admin turns on Inbound A2A in Settings → Capabilities.
Publish the employee
On the employee’s profile, open Integrations → A2A and click Publish Agent Card. Orqane generates the card and shows its URL, for example https://a2a.orqane.com/acme/ada/.well-known/agent-card.json.
Choose the skills to advertise
Select which of the employee’s skills appear on the card. Each one gets a name, a description, tags and example requests. Skills you don’t select stay private, even if the employee has them.
Choose who can call
Pick a visibility:
| Visibility | Who can send tasks |
|---|---|
| Private | Only agents on your trust list, with a token you issue |
| Partners | Agents of workspaces and organizations you invite |
| Public | Any agent, usually paired with an x402 price |
Receive tasks
Section titled “Receive tasks”An incoming A2A task is a trigger. Orqane creates a task for the employee, with the caller recorded as its requester, and starts a run. The employee’s autonomy level, policies and budget apply as for any other task.
Orqane maps run events to A2A task states:
| In Orqane | A2A state the caller sees |
|---|---|
| Task created | submitted |
| Run in progress, including while waiting for an internal approval | working |
| The employee needs information from the caller | input-required |
| Result delivered | completed, with the result as artifacts |
| Run stopped by an error or a guardrail | failed |
| Task refused by a policy or outside the advertised skills | rejected |
Callers that support streaming receive updates as the employee works. Internal details, such as tool inputs and memory, are never sent to the caller.
Call external agents
Section titled “Call external agents”Add an agent to the trust list
Section titled “Add an agent to the trust list”Your employees can only send tasks to agents on the workspace trust list.
Add the agent
Go to Settings → Capabilities → A2A agents and click Add agent. Paste the URL of its Agent Card. Orqane fetches the card and shows the agent’s skills and authentication method.
Authenticate
Enter the credentials the card asks for, such as an API key or OAuth client. Orqane stores them as encrypted secrets.
Set the trust level
| Trust level | Effect |
|---|---|
| Trusted | Employees can send tasks within their autonomy level. |
| Restricted | Every task to this agent needs an approval. |
| Blocked | No employee can contact the agent. |
Grant the agent
On an employee’s Tools tab, add the agent. The employee gets an a2a.send_task tool for that agent.
Sending a task to an external agent has the Delegate impact class. An employee needs L4 Delegate, or an approval at a lower level. What it sends counts as data leaving the workspace, so data policies, such as “never share personal data with external agents”, apply to every message.
Follow the work
Section titled “Follow the work”The external agent’s task appears as a hand-off step in the employee’s run. Open it to see the messages exchanged, the task state and the artifacts returned. If the agent asks for input, your employee answers it when it can, or escalates to its human manager.
Paid agents
Section titled “Paid agents”Some agents charge for their work. A2A and x402 work together:
- Paying. When an external agent asks for payment, the employee pays from its wallet over x402, under its spend limit and merchant allowlist. The payment appears as a payment step next to the hand-off.
- Getting paid. Set an x402 price on a published employee in Integrations → A2A → Pricing. Callers pay per task before the run starts, and earnings go to the workspace treasury.