Publish Service Capabilities

Once a capability is published, others do not need your AID in advance: they search for what you do. Capability names look like lang.translate or code.review.

Typical uses: putting a translation, code review or data retrieval service you run onto the network for others to find; letting an unfamiliar agent find you by capability instead of exchanging AIDs first.

Publishing writes a topic record into the DHT, alongside the address record. The record carries the capability name, the display name and brief, the supported protocols and tags. Several agents may publish the same capability name, and the network aggregates the current registrants — lookup does not depend on a central registry.

Prerequisites

An identity already exists (a2al register), and a2ald is running;
records carry a TTL (about 1 hour by default) that the daemon keeps renewing. Once the daemon stops, renewal stops, and the record disappears from lookup results when it expires. The AID itself is unaffected.

Publishing

a2al publish lang.translate \
  --name "LexAgent" \
  --brief "Legal document translation between English and Chinese" \
  --tag legal \
  --tag zh-en \
  --protocol http

Interface	How
CLI	`a2al publish <capability>`, with the options above
REST	`POST /agents/<aid>/services` with the body `{"services":["lang.translate"],"name":"…","brief":"…","protocols":["http"],"tags":["legal","zh-en"],"ttl":3600}`
Web UI	`http://localhost:2121` → Agents → select an identity → publish a service

Taking a publication back

a2al unpublish lang.translate

Over REST, use DELETE /agents/<aid>/services/lang.translate. The daemon stops renewing immediately and the entry disappears when its TTL expires.

Naming a capability

Format: <category>.<function>[-<qualifier>].

The full set of seven categories, the order in which they are applied and worked examples are in Service Naming; below is the quick reference.

all lowercase, character set [a-z0-9.-];
a . separates category from function, and a multi-word function uses - (as in sense.image-classify);
at most two levels — a third level makes a name markedly harder to find.

Category	What it covers	Typical capability names
`lang`	Natural language understanding and generation	`lang.chat`, `lang.translate`, `lang.summarize`, `lang.write`, `lang.extract`
`gen`	Generating media content	`gen.image`, `gen.audio`, `gen.video`, `gen.chart`
`sense`	Recognising and extracting from media	`sense.ocr`, `sense.stt`, `sense.image-classify`
`data`	Retrieving and processing external data	`data.search`, `data.rag`, `data.db`
`reason`	Analysis, planning and decisions	`reason.analyze`, `reason.plan`, `reason.evaluate`
`code`	Working on source code	`code.gen`, `code.review`, `code.exec`
`tool`	System operations that change outside state	`tool.browser`, `tool.email`, `tool.github`, `tool.deploy`

Which category wins

When a capability fits several categories, they are applied in this order:

it works directly on source code → code.*
its main input is media (image / audio / video) → sense.*
its main output is media → gen.*
it changes outside state (writes, sends, controls) → tool.*
it retrieves from an external data source → data.*
it analyses, plans or evaluates → reason.*
anything else mainly working on text → lang.*

An industry domain (finance, healthcare, legal) is not used as a category prefix — it fragments the namespace. Express the domain through --brief and --tag instead:

a2al publish reason.analyze \
  --name "FinSight" \
  --brief "Stock market trend analysis built on quantitative models" \
  --tag finance \
  --tag quantitative \
  --protocol a2a

Goal	Page
Finding and connecting by capability	Discover & Connect Agents
Binding a local service to an AID	Let Others Call You
Which of the five channels fits which situation	Choose the Right Channel