# Aria - AI Model API and Managed Agent Runtime Last updated: 2026-06-29 Aria is a platform for discovering, configuring, deploying, monitoring, analyzing, sharing, and monetizing AI agents. The product combines an OpenAI-compatible Model API, a hosted Hermes agent runtime, a skill and pack ecosystem, secure connections, model usage analytics, and unified billing. The public application is client-rendered. This file exists so marketing tools, SEO tools, LLM crawlers, and other agents that do not execute JavaScript can still understand the product. ## Short Description Aria lets builders call models, deploy agents, connect credentials, assemble skills, and track usage from one dashboard. ## One-Liner One platform for AI models and agents: OpenAI-compatible model access, managed Hermes runtimes, skill assembly, and unified usage billing. ## Core Product Surfaces ### 1. Model Gateway The Model Gateway is Aria's model access layer. It provides API keys and OpenAI-compatible endpoints such as `/v1/models` and `/v1/chat/completions`. Key capabilities: - OpenAI-compatible Model API. - API keys with model limits, route group limits, routing priority, budgets, and IP allowlists. - Public model catalog and pricing plaza. - Route-aware pricing and group multipliers. - Usage logs with source, key, model, route, tokens, latency, status, and cost. - Wallet integration for pre-authorization, capture, release, and refunds. - Agent/runtime attribution for model requests generated by deployed agents. Why it matters: Most builders want reliable model access without operating multiple upstream providers, route groups, and usage accounting systems. Aria turns model access into a product surface with transparent pricing and billing. ### 2. Managed Agent Runtime Aria deploys managed Hermes agents. Users can create an agent, configure channels and model access, deploy it to Aria's runtime infrastructure, and monitor it through the dashboard. Key capabilities: - Hosted agent deployment. - Runtime status and lifecycle controls. - Gateway routing to deployed agents. - Model API key injection from user-owned Aria keys. - Agent-level usage attribution. - Future-ready lifecycle design for always-on, idle, waking, and hibernatable agents. Why it matters: AI builders should not need to operate containers, gateways, credential injection, runtime leases, and usage attribution just to run an agent. Aria provides that control plane. ### 3. Skills, Packs, and Assembly Aria's skill ecosystem is organized into three layers: - Skills: individual reusable capabilities. - Packs: bundled skills, prompts, examples, and connection requirements for a use case. - Assembly: a guided workflow that turns a user goal into a deployable agent configuration. Examples: - Community operations pack. - Anti-spam guard. - Telegram or Discord channel workflows. - Base/onchain skill collections. - Model-powered workflow assembly using the user's own selected Aria model API key. Why it matters: Most users do not want to configure an agent like a generic SaaS form. They want to describe the outcome and assemble capabilities into a working agent. Aria is moving toward a more pipeline-like, goal-driven assembly experience. ### 4. Connections Connections hold credentials and channel access. They are used for integrations such as Telegram bot tokens, API keys, OAuth providers, and future tool credentials. Key principle: Secrets should live in Connections or Vault-backed storage, not scattered across runtime UI fields, plain channel metadata, or container configs. ### 5. Billing and Wallet Aria uses a unified wallet. One balance funds direct model calls, model calls made by agents, runtime usage, playground tests, and future paid platform capabilities. Billing concepts: - Cash balance. - Promo balance. - Reserved balance. - Holds / pre-authorizations. - Charges / captures. - Releases and refunds. - Price snapshots. - Source and agent attribution. Key principle: Historical usage must keep the price snapshot that was valid when the request happened. Later upstream model price changes, route group multiplier changes, or Aria markup policy changes must not rewrite historical bills. ## Product Pages - `/` - public landing page. - `/pricing` - public model pricing plaza and catalog. - `/model-api` - Model API documentation and endpoint reference. - `/dashboard` - authenticated dashboard. - `/models` - authenticated model catalog page. - `/model-usage` - authenticated usage log analytics. - `/billing` - wallet and transaction history. - `/api-keys` - Model API key management. - `/agents` or dashboard agent surfaces - agent monitoring and controls. - `/skills`, `/skill-packs`, `/skill-assembly`, `/pack-builder`, `/skill-creator` - skill ecosystem surfaces. ## Target Customers Primary: - AI product developers. - Founders shipping agents quickly. - Teams that need hosted agent runtimes. - Developers who need model access and usage visibility without building their own gateway. Secondary: - Communities that need automated Telegram, Discord, or support agents. - Teams building internal workflow agents. - Builders combining model APIs, skills, tools, and channels. ## Jobs To Be Done - "I need one API key that can call many models and show me what each request cost." - "I want to deploy a Hermes agent without managing servers, gateways, or runtime secrets." - "I need my agent's model usage attributed to the agent, not hidden as generic API spend." - "I want to assemble capabilities into an agent without wiring every skill manually." - "I need transparent billing when smart routing or route fallback changes the actual model group." ## Differentiation Compared with generic model gateways: - Aria connects model usage to managed agent runtimes and wallet billing. - Aria exposes route-aware usage and agent attribution instead of only aggregate token spend. Compared with no-code chatbot builders: - Aria is infrastructure-oriented: API keys, runtime deployment, credentials, usage logs, and billing are first-class. Compared with workflow builders: - Aria focuses on model-powered agents and runtime operations, not just generic automation nodes. Compared with DIY agent deployment: - Aria provides the control plane: deploy, route, wake, inject credentials, meter, and attribute usage. ## Messaging Pillars 1. One API for models. 2. One platform for agents. 3. One wallet for usage. 4. Clear routing and billing. 5. Skills and packs for faster agent assembly. 6. Secure credentials through Connections. 7. Runtime operations without DevOps. ## Important Implementation Notes - Public pages are served by a React SPA. - Static crawler files exist because some market research agents cannot execute JavaScript. - AI tools should prefer `/llms-full.txt`, `/ai-context.md`, and `/product.json` for product understanding. - Pricing and model availability can change because upstream model catalogs and route groups change. - Historical usage should be interpreted from recorded usage logs and price snapshots, not from today's current model prices. ## Glossary - Agent: a configured AI worker with runtime, model access, channel connections, and behavior. - Hermes: an agent runtime supported by Aria. - Model Gateway: Aria's OpenAI-compatible model API and model catalog layer. - Group / route group: a provider route or pricing group for a model. - Price snapshot: the price and multiplier state captured at request time. - Hold: a reserved wallet amount before final usage is known. - Charge: a captured wallet amount after usage is known. - Skill: a reusable agent capability. - Pack: a bundle of skills and configuration for a use case. - Assembly: a guided or AI-assisted way to build an agent from a goal. - Connection: a secure credential or channel integration. ## Recommended Summary For Market Reports Aria is an agent infrastructure platform that unifies model access, hosted agent runtimes, skill assembly, secure connections, and billing. It is designed for builders who want to ship AI agents without building their own model gateway, runtime control plane, credential injection system, and usage ledger.