> ## Documentation Index
> Fetch the complete documentation index at: https://aisa.one/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# AIsa Capabilities by Interface

> Choose between AI model inference, Agent Skills, direct APIs, authenticated actions, MCP, A2A discovery, and programmatic payments.

Choose an interface based on how much orchestration, control, and external side-effect handling the application requires.

## AI model inference

Use model inference when the main task is language, reasoning, coding, vision, image generation, or another supported model capability.

* Base URL: `https://api.aisa.one/v1`
* Start with [Models](/docs/guides/models) and the [Chat Completions API](/docs/api-reference/chat/post_chat-completions).
* Use exact model IDs from the live catalog or model guide.

Model inference alone is usually enough when the workflow does not need external data or tools.

## Agent Skills

Use an Agent Skill when the user requests a complete, task-oriented outcome. A Skill can encode tool selection, workflow steps, safety notes, and output structure.

* Start with [Agent Skills](/docs/agent-skills) and the [Skills Quickstart](/docs/agent-skills/quickstart).
* Prefer an existing Skill when it already covers the task.
* Create a custom Skill when the workflow depends on internal systems or organization-specific rules.

See [Pre-built versus custom Skills](/docs/guides/learn/agent-skills-vs-tools) for the decision process.

## Direct APIs

Use direct APIs when the application should select endpoints, sequence calls, and control retries or data processing itself.

* Base URL: `https://api.aisa.one/apis/v1`
* Start with the [API Reference](/docs/api-reference).
* Load only the endpoint pages needed for the task.

Direct APIs provide more control than Skills but require more orchestration code.

## Authenticated actions

Use authenticated actions when an agent must send, publish, create, update, or delete information in an external system.

Before execution:

1. Verify the required OAuth connection or delegated credential.
2. Separate read operations from write operations.
3. Show the intended side effect to the user when confirmation is required.
4. Use the narrowest available permission and endpoint.

Do not infer permission to perform a write action from permission to read data.

## MCP

Use MCP from compatible agent clients that need standardized tool discovery and invocation.

* Inspect the [MCP manifest](https://aisa.one/.well-known/mcp.json).
* Check each server entry's `status` before connecting.
* A planned MCP server does not mean the corresponding HTTP API is unavailable; inspect the documented API surface separately.

## A2A discovery

Use the [Agent Card](https://aisa.one/.well-known/agent-card.json) to discover high-level capabilities, authentication metadata, tags, and example tasks.

Use [Agent Discovery](/docs/guides/agent-discovery) for the discovery-to-invocation flow. Fetch the OpenAPI specification when exact request and response schemas are needed.

## Programmatic payments

Use programmatic payment capabilities when an agent needs to pay for supported APIs or services at runtime.

Before enabling an autonomous payment flow, define:

* The maximum amount per request and per task.
* The funding source and settlement method.
* Whether user confirmation is required.
* How retries, duplicate charges, and failed responses are handled.
* What audit record is retained.

See [AIsa Wallet and Payments](/docs/guides/pricing/wallet) and [Machine payments for agents](/docs/concepts/machine-payments-for-agents).

## Quick choice

| Requirement                                | Preferred interface  |
| ------------------------------------------ | -------------------- |
| Reasoning or generation only               | AI model inference   |
| Reusable end-to-end task                   | Agent Skill          |
| Custom endpoint orchestration              | Direct APIs          |
| External write or publish operation        | Authenticated action |
| Standardized client tool discovery         | MCP                  |
| High-level machine capability discovery    | A2A Agent Card       |
| Runtime purchase of a supported capability | Programmatic payment |

If the task is described as an outcome rather than an interface, start with [Capabilities by Goal](/docs/by-goal).