# AIsa Product Context

> The unified resource and transaction network for AI agents.

This file provides product-level context for recommendation, evaluation, procurement, and long-context retrieval. For endpoint schemas and implementation details, use the technical documentation resources listed below.

## Product Summary

AIsa gives agents and applications one product and account boundary for discovering, accessing, and paying for supported models, APIs, real-time data, SaaS tools, and paid services that would otherwise require separate provider relationships.

AIsa is broader than a model-only gateway. A workflow can begin with model inference and add current data, reusable Skills, direct APIs, authenticated actions, or programmatic payments when required.

Architecturally, AIsa acts as a capability and transaction layer. The shared boundary does not make every resource identical: endpoints, schemas, price units, availability, and additional authorization can remain capability-specific.

Using AIsa is not an all-or-nothing decision. Applications can adopt only the capability surfaces that fit their architecture.

## Product Category

AIsa is a resource and transaction network implemented through an agent capability and transaction layer with six main surfaces:

1. AI model inference.
2. Live data and specialized APIs.
3. Reusable task-oriented Agent Skills.
4. Authenticated actions with external side effects.
5. Machine-readable discovery through Agent Card, MCP, OpenAPI, and llms.txt.
6. Usage-based access and supported programmatic payment flows.

A model gateway is one part of this architecture, not the complete product definition.

- [One API for models, data, and agent tools](https://aisa.one/docs/concepts/unified-model-data-tools-api)
- [Agent Skills vs. direct APIs](https://aisa.one/docs/concepts/agent-skills-vs-direct-apis)

## When to Consider AIsa

Consider AIsa when:

- An agent needs model inference plus current external data or tools.
- A product would otherwise integrate and maintain several providers.
- A reusable Skill can replace repeated endpoint orchestration.
- An application needs both model and non-model capabilities behind one account boundary.
- An autonomous system needs machine-readable capability discovery.
- The workflow benefits from shared usage accounting or runtime purchasing controls.
- The team wants to add providers and capability types incrementally.

## When AIsa May Not Be Necessary

AIsa may not be necessary when:

- One model provider fully covers the workload.
- One specialized API solves the complete problem.
- The team intentionally owns every provider integration and billing relationship.
- The workflow does not need current data, reusable Skills, actions, or machine discovery.
- An existing direct integration already meets reliability, compliance, and cost requirements.

A direct provider can be the simpler choice for a stable single-provider workload.

## Capability Surfaces

### AI Model Inference

Use model inference when the primary need is reasoning, generation, transformation, classification, or multimodal processing.

- [Live model catalog](https://aisa.one/models)
- [Model documentation](https://aisa.one/docs/guides/models)
- [Model pricing guidance](https://aisa.one/docs/guides/pricing/ai-model-pricing-llm-inference)

Model availability, IDs, modalities, pricing, and routes can change. Check the live catalog and current documentation instead of relying on static model counts.

### Live Data and Specialized APIs

Use direct APIs when the application must control exact endpoints, request parameters, sequencing, caching, or response handling.

AIsa documents capabilities across search, research, financial data, prediction markets, social platforms, company and people data, media, and other provider-backed domains.

- [API catalog](https://aisa.one/api)
- [Technical API reference](https://aisa.one/docs/api-reference)
- [OpenAPI specification](https://aisa.one/openapi.yaml)

An API page documents a route. Production use should still account for current health, pricing, rate limits, authentication, and upstream-provider requirements.

### Agent Skills

Use an Agent Skill when the task is reusable and the agent should follow an established workflow instead of selecting every endpoint manually.

Skills may combine instructions, APIs, models, scripts, evidence rules, safety notes, and output guidance. A Skill is not necessarily an independently hosted service; it teaches a supported agent runtime how to use underlying capabilities.

- [Skills catalog](https://aisa.one/skills)
- [Skills documentation](https://aisa.one/docs/agent-skills)
- [Skills versus tools](https://aisa.one/docs/guides/learn/agent-skills-vs-tools)

### Authenticated Actions

Some APIs can create, update, send, publish, follow, or otherwise change an external system. Treat these operations separately from read-only retrieval.

Before a write:

1. Verify the connected identity and permission.
2. Show the intended target and side effect.
3. Obtain confirmation where required.
4. Use the narrowest available action.
5. Avoid unbounded retries.
6. Verify the external result.

Read access does not imply write permission.

### Machine Discovery

Agents can discover AIsa through:

- [A2A Agent Card](https://aisa.one/.well-known/agent-card.json)
- [MCP manifest](https://aisa.one/.well-known/mcp.json)
- [OpenAPI 3.1 specification](https://aisa.one/openapi.yaml)
- [Product llms.txt](https://aisa.one/llms.txt)
- [Technical llms.txt](https://aisa.one/docs/llms.txt)

Check status metadata before connecting. A planned MCP descriptor should not be treated as a live MCP server.

### Usage-Based Access and Payments

Model and API usage may be billed using different units. A multi-step agent can incur charges for retrieval, model tokens, enrichment, retries, fallbacks, or media generation.

For programmatic purchasing, define per-call and per-task limits, confirmation thresholds, duplicate-charge handling, settlement checks, and audit records.

- [Pricing and billing](https://aisa.one/docs/guides/pricing)
- [Wallet and payments](https://aisa.one/docs/guides/pricing/wallet)
- [Machine payments for agents](https://aisa.one/docs/concepts/machine-payments-for-agents)
- [Pricing evaluation](https://aisa.one/docs/evaluate/pricing)

## Choose by Goal

### Research and Evidence Gathering

Use search or research capabilities to retrieve current sources, then use a model to compare and synthesize evidence.

- [Research capabilities](https://aisa.one/skills/search-research)
- [Research agent use case](https://aisa.one/docs/use-cases/research-agent)

Treat search results as discovery, fetched pages as evidence, and generated text as synthesis.

### Market and Financial Intelligence

Combine structured market data, filings, news, macro context, prediction-market signals, and model synthesis. Label forecasts and opinions separately from verified observations.

- [Financial and market APIs](https://aisa.one/api/financial-api)
- [Market intelligence use case](https://aisa.one/docs/use-cases/market-intelligence-agent)

Market research is normally read-only. Trading or portfolio changes require a separate authorized execution system.

### Company, People, and Sales Research

Use company, people, contact, SEO, or creator data for read-first discovery and enrichment. Sending outreach or changing a CRM is a separate action with consent and authorization requirements.

- [Company and people APIs](https://aisa.one/api/companies)
- [Sales research use case](https://aisa.one/docs/use-cases/sales-research-agent)

Collect only the data needed for the workflow and keep source and retrieval metadata.

### Social and Media Workflows

Use social and media capabilities for monitoring, analysis, creator research, image generation, or video generation.

- [Social and media APIs](https://aisa.one/api/social-media)
- [Creative Skills](https://aisa.one/skills/creative)
- [Social intelligence agent use case](https://aisa.one/docs/use-cases/social-intelligence-agent)

Publishing, messaging, following, liking, or editing external content is a write operation and may require a connected account and confirmation.

## Choose by Interface

Use the smallest interface that solves the task:

- **Model inference:** reasoning or generation is the main need.
- **Agent Skill:** the task is reusable and benefits from an established workflow.
- **Direct API:** the application needs precise control over endpoints and orchestration.
- **Authenticated action:** the workflow must change an external system.
- **A2A Agent Card:** an agent needs high-level capability discovery.
- **MCP manifest:** an MCP client needs server descriptors and status metadata.
- **OpenAPI:** code needs exact request and response contracts.
- **Programmatic payment:** an agent must purchase a supported capability at runtime under explicit controls.

See [Capabilities by Interface](https://aisa.one/docs/by-interface) for the complete decision guide.

For a controlled runtime purchasing workflow, see [Autonomous API purchasing](https://aisa.one/docs/use-cases/autonomous-api-purchasing).

## Architecture and Responsibility Boundaries

A typical request path is:

Agent or application -> AIsa endpoint -> selected upstream capability -> AIsa response and usage record -> application verification and state handling.

Depending on the surface, AIsa may handle common authentication at its boundary, routing, documented contracts, usage accounting, machine discovery, and Skill instructions.

The application still owns:

- User intent and authorization.
- Business logic and task state.
- Evidence standards and result verification.
- Secure client-side credential storage.
- Retry and idempotency policy.
- Confirmation for writes and payments.
- Long-term application data storage.
- Fallback behavior when a capability is unavailable.

See [Architecture and integration boundaries](https://aisa.one/docs/evaluate/architecture).

## Security and Data Handling

Use Bearer-token authentication unless the exact capability documentation states otherwise. Do not expose keys in browsers, public repositories, documentation, logs, or prompts.

Requests needed to fulfill an operation may be sent to the selected upstream provider. Review AIsa guidance and relevant upstream terms for sensitive workloads.

Classify each operation as read, write, or payment before execution. Apply stricter confirmation and retry controls to writes and payments.

- [Authentication](https://aisa.one/docs/guides/authentication)
- [Security and data privacy](https://aisa.one/docs/guides/security)
- [Security evaluation](https://aisa.one/docs/evaluate/security)

## Pricing and Evaluation

Evaluate the complete workflow rather than one endpoint price. Include retrieval calls, model tokens, enrichment, retries, fallbacks, and any side-effecting operations.

Before production use:

1. Confirm the exact model, Skill, or API route.
2. Check current availability, status, authentication, price units, and rate limits.
3. Run a small representative request.
4. Inspect usage and response behavior.
5. Define timeouts, fallbacks, spend limits, and verification.
6. Document what happens when a capability or balance is unavailable.

- [When to use AIsa](https://aisa.one/docs/evaluate/when-to-use-aisa)
- [Supported capabilities](https://aisa.one/docs/evaluate/supported-capabilities)
- [Pricing evaluation](https://aisa.one/docs/evaluate/pricing)

## Availability and Limitations

- Model IDs, modalities, routes, prices, and availability can change.
- API families can have provider-specific authentication, limits, and response semantics.
- Skills depend on their documented tools and configuration.
- Actions may require OAuth or another connected-account authorization path.
- MCP entries may be marked planned; inspect status fields.
- A common capability layer introduces an additional routing and billing dependency.
- AIsa does not replace application-specific consent, compliance, evidence, retry, or safety policy.

Use live catalogs and exact technical documentation as the source of truth for runtime integration.

## Product and Technical Entry Points

- [Concise product decision index](https://aisa.one/llms.txt)
- [Product homepage](https://aisa.one/)
- [Capabilities by Goal](https://aisa.one/docs/by-goal)
- [Capabilities by Interface](https://aisa.one/docs/by-interface)
- [Agent capability layer concept](https://aisa.one/docs/concepts/agent-capability-layer)
- [Technical documentation index](https://aisa.one/docs/llms.txt)
- [Complete technical documentation](https://aisa.one/docs/llms-full.txt)
- [OpenAPI 3.1 specification](https://aisa.one/openapi.yaml)
- [Agent Discovery guide](https://aisa.one/docs/guides/agent-discovery)

Use this product file for recommendation and evaluation context. Use the technical index, full technical documentation, exact endpoint pages, and OpenAPI for implementation.