Capability Discovery

ModelMesh uses a hierarchical capability system to route requests. Instead of memorizing full dotted paths like generation.text-generation.chat-completion, you can use short aliases and the discovery API. For the complete hierarchy tree, see Model Capabilities. For how capabilities map to pools and routing, see the architecture overview.

Capability Aliases

Alias	Full Path
`chat-completion`	`generation.text-generation.chat-completion`
`text-generation`	`generation.text-generation`
`code-generation`	`generation.text-generation.code-generation`
`text-embeddings`	`representation.embeddings.text-embeddings`
`text-to-speech`	`generation.audio.text-to-speech`
`speech-to-text`	`understanding.audio.speech-to-text`
`text-to-image`	`generation.image.text-to-image`
`image-to-text`	`representation.image.image-to-text`

Discovery API

Python

import modelmesh

# List all aliases
caps = modelmesh.capabilities.list_all()
# ['chat-completion', 'code-generation', 'image-to-text', ...]

# Resolve alias → full path
path = modelmesh.capabilities.resolve("chat-completion")
# 'generation.text-generation.chat-completion'

# Dotted paths pass through unchanged
modelmesh.capabilities.resolve("generation.text-generation")
# 'generation.text-generation'

# Unknown aliases return unchanged
modelmesh.capabilities.resolve("custom-cap")
# 'custom-cap'

# Search by keyword (case-insensitive)
modelmesh.capabilities.search("text")
# ['text-embeddings', 'text-generation', 'text-to-image', 'text-to-speech']

# View the hierarchy tree
tree = modelmesh.capabilities.tree()
# {
#   'generation': {
#     'text-generation': {
#       'chat-completion': {},
#       'code-generation': {},
#     },
#     'audio': {'text-to-speech': {}},
#     'image': {'text-to-image': {}},
#   },
#   'representation': { ... },
#   'understanding': { ... },
# }

TypeScript

import * as capabilities from '@nistrapa/modelmesh-core/capabilities';

const caps = capabilities.listAll();
const path = capabilities.resolve('chat-completion');
const matches = capabilities.search('text');
const tree = capabilities.tree();

Using Capabilities

When creating a client, the capability name determines which pool handles your requests:

# These are equivalent:
client = modelmesh.create("chat-completion")
client = modelmesh.create("generation.text-generation.chat-completion")

When calling create(), ModelMesh resolves the capability to find or create a pool containing all models that support it.

Hierarchy

The capability tree follows a three-level hierarchy:

Category
└── Domain
    └── Specific Capability

Categories:

generation — Create new content (text, audio, images)
representation — Transform content into structured forms (embeddings, descriptions)
understanding — Analyze and interpret content (transcription, classification)

Custom Capabilities

You can register custom capabilities in your YAML configuration:

models:
  my-model:
    provider: my-provider
    capabilities:
      - generation.custom.my-capability

pools:
  my-pool:
    capability: generation.custom.my-capability

Custom capability paths don’t need aliases — use the full dotted path directly.

See also: FAQ · Model Capabilities · System Configuration