Capabilities

What work Skilleo can perform. Callers request versioned capabilities (image.generate@1), not model names.

Skilleo routes each capability to a connected provider. You manage provider keys; Skilleo stores connection pointers only. See Providers and Platform availability.

Agentic apps should prefer runs (POST /v1/runs). Infrastructure apps may still map to jobs.

Routing policy

Capability	Allowed connections	Rule
llm.generate	OpenAI, Anthropic, Gemini, MiniMax, AWS, RunPod	Any connected LLM provider
speech.generate, voice.clone	MiniMax, ElevenLabs	One of — no other voice backends
image.generate, GPU	RunPod	RunPod only

GET /v1/capabilities

List available capabilities.

curl -s https://api.skilleo.org/v1/capabilities \
  -H "Authorization: Bearer $SKILLEO_API_KEY"

Response data

Field	Type	Description
items	array	Capability objects
count	integer	Length of items

Each capability includes (additive discovery fields):

Field	Description
id	Stable identifier (e.g. image.generate)
name	Same as id for discovery
title	Human label
version	Capability version string (e.g. "1")
description	What the capability does
status	Capability readiness (available, or planned when documented ahead of launch)
source.providers[].status	Per-backend readiness in the registry
async	Whether execution is asynchronous
estimated_cost	Whether provider usage may incur cost
input_schema / output_schema	JSON-schema style shapes
invoke	Preferred invoke: POST /v1/runs
job	Alternative: POST /v1/jobs with type
source.providers	Backends and requires_connection ids
source.routing	allowed_connections, policy, default_provider
input / output / targets	Older field docs (still present)

GET /v1/capabilities/{id}

Return one capability (e.g. /v1/capabilities/image.generate or image.generate@1).

curl -s https://api.skilleo.org/v1/capabilities/llm.generate \
  -H "Authorization: Bearer $SKILLEO_API_KEY"

Code	When
NOT_FOUND	Unknown id or unsupported version

Registered capabilities

Id	Version	Status	Routing	Notes
image.generate	1	available	RunPod only	GPU pod or serverless endpoint
research.gather	1	available	—	Wikipedia (CPU); optional LLM synthesis when connected
composition.layout	1	available	—	CPU social card layout
llm.generate	1	available	Any LLM; default minimax	MiniMax, OpenAI, Anthropic, Gemini, Bedrock, RunPod
text.generate	1	available	same as llm.generate	Alias for editorial copy
speech.generate	1	available	MiniMax or ElevenLabs; default minimax	MP3 output
voice.clone	1	available	MiniMax or ElevenLabs; default minimax	Reference audio required

Connect a provider before calling a capability that requires one. Without a matching connection you get CREDENTIALS_MISSING. Pin a backend with options.provider when you have several LLM or voice connections.

Example: LLM — any connected provider

curl -s -X POST https://api.skilleo.org/v1/runs \
  -H "Authorization: Bearer $SKILLEO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"target":"capability","name":"llm.generate","version":"1","input":{"objective":"Write a hook about Kenya wildlife"},"options":{"provider":"openai"}}'

from skilleo import Client

client.providers.connect("openai")

run = client.capabilities.run(
    "llm.generate",
    version="1",
    input={"objective": "Write a hook about Kenya wildlife"},
    options={"provider": "openai"},
)
done = client.runs.wait(run["id"])
print(done["data"]["output"])

Example: voice — MiniMax or ElevenLabs

curl -s -X POST https://api.skilleo.org/v1/runs \
  -H "Authorization: Bearer $SKILLEO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"target":"capability","name":"speech.generate","version":"1","input":{"text":"Earth stops spinning."},"options":{"provider":"elevenlabs"}}'

client.providers.connect("elevenlabs")

run = client.capabilities.run(
    "speech.generate",
    version="1",
    input={"text": "Earth stops spinning."},
    options={"provider": "elevenlabs"},
)
client.runs.wait(run["id"])

Example: image — RunPod only

curl -s -X POST https://api.skilleo.org/v1/runs \
  -H "Authorization: Bearer $SKILLEO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"target":"capability","name":"image.generate","version":"1","input":{"prompt":"a red cube","steps":8},"options":{"endpoint_id":"ep_..."}}'

client.providers.connect("runpod")

run = client.runs.create(
    name="image.generate",
    version="1",
    input={"prompt": "a red cube", "steps": 8},
    options={"endpoint_id": "ep_..."},
)
client.runs.wait(run["id"])

Infrastructure apps may still use client.jobs.create(type="image", ...) — see Jobs.