A2A remote agents 

nr-llm can hand a message to a remote agent that speaks the Agent2Agent protocol (A2A) 1.0 over JSON-RPC, and follow the task the remote agent creates. This page covers the client: a record per remote agent, a PHP API, the commands, agent runs that delegate a step to a remote agent and wait for it (ADR-281, see Delegating a step of an agent run), and push notifications that shorten the wait.

Register a remote agent 

Create an A2A Remote Agent record (list module, any page or the root). The record holds:

Identifier, title, description
The identifier names the agent on the command line.
Agent card URL
The URL of the remote agent card, usually ending in /.well-known/agent-card.json.
Credential
Stored in nr-vault, never in the record. nr-vault adds it as Authorization: Bearer to every request to this agent, and writes its audit row. Leave it empty for an agent without authentication. The card itself is fetched without the credential. A credential is only sent over https: a request with a credential to an http URL is refused before anything leaves (CREDENTIAL_NEEDS_HTTPS).
Data class
What may be sent to this agent. There is no default: a record without a data class does nothing, the card import included.
Requires approval
On by default. A run that calls this agent waits for a person's approval first (ADR-084).
Enabled
Off by default. Nothing is sent to a disabled agent.
Limits
Maximum response size (default 1 MiB; a larger response is not parsed. The transfer stops at the limit when the nr-vault client streams, which needs PHP's curl_multi_* functions; otherwise the whole body is received and the request timeout bounds the transfer), request timeout (default 30 s), connect timeout (default 5 s; see Known limits), and the limits of a delegating run: maximum wait (default 3600 s), minimum and maximum poll interval (5 s and 300 s) and delegations per run (3).

Nothing the remote agent's card says sets the data class, the enabled flag or the approval flag: the remote does not decide its own trust.

Import the agent card 

Import or re-import the agent card
vendor/bin/typo3 nrllm:a2a:import <identifier>
Copied!

The import fetches the card through the address guard, checks it against the A2A 1.0 proto and selects the first interface with binding JSONRPC and protocol version 1.0. A card that offers no such interface is refused, and the message names what it offers instead. When the interface names a tenant, every request carries it.

The interface URL is where the credential goes. When it is not on the card URL's own scheme, host and port, the first import already marks the record Card changed, confirm it: a card cannot move the credential to another host, or from https to http, without an administrator seeing it.

Import again whenever the remote agent changes. If the new card has another interface URL, another tenant or other capabilities, the record shows Card changed, confirm it, and the client refuses the agent until an administrator has looked at the card, chosen Confirm: <interface URL> in Confirm the card and saved. The choice names the card the form showed: if an import changed the card while the form was open, the save confirms nothing and the record log says to review the card again. Saving the record without the choice confirms nothing, and the command-line user cannot confirm. The confirmed interface URL is shown on the record; an interface off the card URL's host is used only while it is that one.

Changing the card URL or the credential, including entering a new secret value for the same credential, drops the imported card: import it again, and confirm it if asked, before the agent is used. A secret replaced in the nr-vault module, outside this record, is not noticed.

Only administrators can open or edit remote agent records.

Which hosts the client may reach 

Every request, the card import included, goes through the address guard of ADR-202 under its own egress group a2a_agent (ADR-279): http or https, no host of this installation, every address the host resolves to public, the connection pinned to exactly the addresses that were checked, no redirect followed, nothing retried.

The A2A client has host lists of its own in the extension configuration, separate from those of the fetch_external_url tool:

a2a.client.allowedHosts
Empty (default) allows every public host. A host on a port other than 80 or 443 needs an entry with the port, for example agent.example.org:8443.
a2a.client.deniedHosts
Hosts and addresses that are always refused.
a2a.client.allowViaProxy
Behind an HTTP proxy the address checks cannot reach past the proxy, so A2A requests are refused unless this is on and the allowlist is non-empty.

A remote agent on a private, loopback or link-local address cannot be reached; no list entry opens that.

Send a message from the command line 

The commands act as the backend user in the extension setting worker.technicalBeUserUid: nr-vault reads the credential as that user, and the user owns the tasks the commands create. Without one configured the commands refuse.

Send a message, then read the task
vendor/bin/typo3 nrllm:a2a:send <identifier> "Summarise the release notes"
# prints the remote task id and exits

vendor/bin/typo3 nrllm:a2a:get <identifier> <task-id>
# prints the task: state, status message, artifacts
Copied!

nrllm:a2a:send sends with returnImmediately, so it returns as soon as the remote agent has accepted the task. nrllm:a2a:get reads it later, as often as needed. Only tasks this installation sent can be read. Text from the remote agent is printed as plain text: console markup is shown literally, and control characters (C0 and C1) and format characters such as bidi overrides are removed, the line break excepted.

What happens to a message 

Before a message leaves, the client writes an intent to the table tx_nrllm_a2a_send and commits it: owner, agent, message id and the sha256 of the request. The intent then ends in exactly one state:

accepted
The remote agent created a task. Its task and context ids are stored, and the row is the local handle of the task.
answered
The remote agent answered with a message and created no task.
refused
The remote agent answered with a JSON-RPC error (code and reason are stored), or the client refused locally: the guard refused the address, the credential is missing, a redirect, an invalid response with a status other than 2xx or 5xx.
outcome_unknown
The request may have reached the remote agent and no response was read: a timeout after sending, a lost connection, a response larger than the record's limit (RESPONSE_TOO_LARGE), a 5xx without a JSON-RPC body, a 2xx that is not a valid response (INVALID_AGENT_RESPONSE), a response of a kind A2A 1.0 does not know.

An unknown outcome is never sent again, by the client or anything else. The remote may well have created a task, but no task id came back, so nrllm:a2a:get and listTasks() cannot find it: they know only tasks with a local handle. Check on the remote agent's side, with its operator or its own tools, whether the message id created a task before sending again; a resend is a new message with a new message id.

The same owner sending the same request under the same message id again gets the first intent back and sends nothing. The same message id with a different request is refused before anything is sent.

PHP API 

Netresearch\NrLlm\Service\Agent\A2a\Client\A2aClient is a public service:

Send a message and read the task from PHP
$outcome = $client->sendMessage(
    $agent,                       // A2aRemoteAgent record
    $backendUserUid,              // the local actor, owner of the task
    A2aMessage::userText(A2aClient::newMessageId(), 'Hello'),
    A2aSendMessageConfiguration::returningImmediately(),
);

if ($outcome->result?->isTask()) {
    $taskId = $outcome->intent->remoteTaskId;
    $task   = $client->getTask($agent, $backendUserUid, $taskId);
}
Copied!

sendMessage() returns an A2aSendOutcome: the journal row, whether this call sent anything, and the result, which is a task or a message. A caller has to handle both. getTask() and cancelTask() work only on a task the same backend user created; another user's task reads as not found, and the remote agent is not asked. listTasks() returns only the caller's tasks, even when several users share one remote credential, and creates no handle.

The classes are @internal while the A2A stages land; their names and signatures may still change.

Delegating a step of an agent run 

Each enabled remote agent with an imported card offers one tool per skill on its card, named a2a_<identifier>_<skill id>. Like an MCP tool it is off until an administrator enables it on a configuration, it carries the record's data class (ADR-094), it asks for approval when the record says so, and it goes through the egress group a2a_agent. Listing the tools opens no connection. Approval, data class and the enabled switch are read from the record when the tool is called, so a change reaches a running worker.

A model turn that calls such a tool together with any other tool is refused before anything of it runs; the model is told to call the delegation alone. At most one delegation is pending per run segment.

The tool sends the message with returnImmediately. A message answer, or a task the remote agent already finished, is the tool's result and the run goes on. Otherwise the run is suspended in the status waiting_for_dependency: it holds no lease and no worker, the inbox shows it in the section Waiting for a remote agent, never as a decision, and nrllm:agent:status counts it.

Schedule the poller like nrllm:agent:work, at least as often as the shortest minimum poll interval of the records:

vendor/bin/typo3 nrllm:a2a:poll
Copied!

It looks at each waiting run's remote task first at the record's minimum poll interval after the send, then at twice the last interval up to the maximum, with up to a quarter of jitter. A task that finished or failed continues the run with the remote's answer as the tool's result, as the run's owner and in its workspace. A request that fails on the way counts as no news. After the record's maximum wait the run continues with a "timed out" result and the remote task is cancelled once. The poller acts as worker.technicalBeUserUid towards nr-vault, like the commands.

When the remote agent asks a question 

A remote task in TASK_STATE_INPUT_REQUIRED asks for something. The model is not asked; the run waits for the person who started it (ADR-311):

  • The run moves to the waiting runs with an input form, whose field shows the remote agent's question, marked as the remote agent's.
  • Only the backend user who started the run can answer it. An administrator or a user with the approval grant who did not start it is refused.
  • The answer goes to exactly the remote task that asked, with its task and context id, through the journal and with secret shapes masked. It is not approved again. Then the run waits for that task again, with a fresh maximum wait; a further question asks again.
  • The maximum wait of the delegation also runs while nobody answers: at the deadline the run continues with the "timed out" result and the remote task is cancelled once. A run cancelled or failed while it waits for the answer has its remote task cancelled by the next nrllm:a2a:poll.
  • An answer is not a delegation: the record's delegations per run count only the first send of each delegation.

A remote task in TASK_STATE_AUTH_REQUIRED asks for authentication, which this installation never gives. Nothing is sent in reply; the poller keeps looking until the task goes on or the maximum wait has passed, and the "timed out" result then says that the remote asked for authentication.

Cancelling a waiting run (nrllm:agent:cancel) cancels its remote task once, as the owner of the delegation, and the run ends cancelled whatever the remote answers; the answer is kept in the journal (cancel_outcome).

The limits of a run (its round cap, token budget, and the delegations it made to each agent) carry over every wait. The delegation limit counts only what this installation sends; a remote agent that delegates further does so outside nr-llm's control.

Push notifications 

Push is off by default. With Receive A2A push notifications (a2a.client.push) on, a delegation asks the remote agent to post the changes of its task to this installation, so the run does not wait for the next poll (ADR-285). Polling stays: a callback only brings the next GetTask forward.

  • Endpoint. Site of the A2A push endpoint (a2a.client.pushSite) names the site whose A2A base URL (site setting nrllm.a2a.baseUrl, see A2A server) carries POST <base>/push. The worker that registers a callback has no site of its own, so the site is named here. The endpoint answers whether or not the server role is on. With push off, or without such a site, it does not exist: the request passes through to TYPO3 like any other URL, as the A2A paths do while the server role is off.
  • Registration. Only a remote whose card declares pushNotifications is asked. After the remote accepted the task, nr-llm writes a local registration with a new random callback credential, sends CreateTaskPushNotificationConfig with the callback URL and the credential as authentication (scheme Bearer, no token), stores the config id the remote answers, and then reads the task once with GetTask, so a change between the send and the registration is not missed. A task settled by then is the tool's result at once. The credential is shown to the remote only; nr-llm stores its public prefix and its SHA-256 hash, as for the tokens of A2A clients (ADR-280). A refused create leaves the run to polling. A create whose outcome is unknown (a timeout after it was sent) is never sent again; the credential stays valid, because the remote may hold the config.
  • Callbacks. A callback needs Authorization: Bearer with the credential of an open registration; anything else is 401 with WWW-Authenticate: Bearer realm="a2a-push" and stores nothing. The answer is the same for a missing, unknown or wrong credential, for a closed registration, and for a delegation to a remote agent record that is disabled or deleted; enabling the record again accepts its callbacks again. An X-A2A-Notification-Token header is ignored: it neither authenticates nor refuses. A2A-Version and a content type are not required. Failed attempts are logged with the source address, never with the credential, and limited per address by the rate limiter nrllm-a2a-push-auth (10 per 15 minutes; override it in $GLOBALS['TYPO3_CONF_VARS']['SYS']['rateLimiter']['nrllm-a2a-push-auth']).
  • Inbox. A body above a2a.client.maxPushBytes (default 64 KiB) is 413, and a declared Content-Length above it is refused before the body is read. A body that is not a JSON object is 400. A valid callback is stored sealed, like the run state, and only then answered 202, which means received and nothing more. The inbox holds at most a2a.client.maxPushInboxPerTask (default 8) callbacks per delegation and a2a.client.maxPushInbox (default 1000) in total that the worker has not looked at yet; one more is 429 with Retry-After.
  • What a callback does. nrllm:a2a:poll looks at a delegation with a stored callback at the record's minimum poll interval after its last look instead of at the scheduled time, never earlier. The look is GetTask as the run's owner; the state the run continues with is what that answers, never the callback body. A look deletes every callback received until it began, so the inbox bounds also bound what is stored; a callback received during the look stays and brings the next look forward. When the run continues, its registration and its callbacks are deleted; a later callback is 401, and a duplicate never continues a run twice.
  • Cleanup. A registration takes callbacks only while its run is running or waiting for the remote task. Registrations and their callbacks are deleted when the run continues (with the remote's result or at its deadline, also when the run then waits for another delegation) or is cancelled, by nrllm:a2a:poll once the run has ended any other way (it failed, was reaped), and with their run by nrllm:privacy:purge. The remote's config is not deleted; a sender that keeps posting gets 401.

Known limits 

  • Connect timeout. nr-vault takes one timeout per request. The record's request timeout bounds the whole request; the connect timeout is bounded by it and by TYPO3's HTTP.connect_timeout, not on its own.
  • Protocol. A2A 1.0 over JSON-RPC only. A card that offers only gRPC, HTTP+JSON or protocol 0.3 cannot be imported. Fields of later protocol versions are ignored; an unknown task state never counts as completed.
  • Streaming is for the PHP API and the command line, not for runs. A run that delegates sends SendMessage and is resumed by polling, so it never holds a worker for the remote's work (ADR-281). A2aClient::sendStreamingMessage(), A2aClient::subscribeToTask() and nrllm:a2a:send --stream read an agent's stream when its card declares streaming: the remote task is stored with the first event that names it, a stream that breaks before that is an unknown outcome and is never sent again, and a stream that ends before the task finished leaves it to GetTask (nrllm:a2a:get). A stream lasts at most the record's maximum wait, at least its request timeout; one event may be at most the record's maximum response size.
  • Push without an embedded config. nr-llm registers a callback with CreateTaskPushNotificationConfig after the remote accepted the task, never inside SendMessage. A callback the remote sends before that request is impossible, because the remote does not know the URL yet.
  • The total inbox bound is checked in the transaction that stores the callback, under a lock on the registration only. Callbacks of different delegations arriving at the same moment can exceed it by their number; the bound per delegation is exact.
  • No files. The client sends text only, and of a remote result it reads the text parts; a raw, url or data part of an artifact is not passed on to the run. Files in the client role are part of spec 051 stage 5c and not built yet; the server role takes inbound files, see Files.
  • A remote question is answered by the run's initiator only, as text. A run started by an A2A client of this installation could pass the question on to that client, but the delegation tools need an administrator, and a client's technical user usually is not one. A request for authentication is never answered (see When the remote agent asks a question).
  • A run that cannot be suspended after its message was accepted fails; the remote task stays in the journal as accepted and is not sent again.