Riven · updated 2026-11-16

The AI Platform: Search, Models, Agents and API on One Bill

An AI platform is the layer that turns a collection of AI point tools into one account: a model fleet, a search and research capability, agent execution with audit, a code workspace, an API and a deployment option, all metered on one bill. The 2026 market splits into search engines, assistants, model APIs and suites; buying three or four of them is normal and expensive. Riven is built as one platform - seven Riven-owned lanes, a Research lane with citations, Control and Forge operators on a Propose-Approve-Act-Audit rail, code, an AI browser and an OpenAI-compatible API at a flat $3.00 per 1M tokens - from $15 per month, with self-hosted single-tenant deployment. This page defines the layers, scores the market against them and shows the pricing.

What makes something an AI platform

The word platform is overused, so it is worth defining narrowly. A tool becomes a platform when the same account, identity and billing cover several distinct AI jobs, and when developers can reach the same capability through an API rather than only through a user interface. By that test there are four layers that matter in 2026.

  • The model layer: which models you can call, whether they are first-party or routed, and how per-task routing is priced.
  • The retrieval layer: whether the platform can search and cite, how fresh the index is, and whether research is a first-class capability or an add-on.
  • The execution layer: agents and code that can act, with an audit trail and an approval rail rather than an opaque loop.
  • The delivery layer: an API with predictable economics, plus a deployment option for teams whose data cannot leave their infrastructure.

How the 2026 market maps onto those layers

Search engines are strong on retrieval and weak on execution and delivery. Perplexity, for instance, has excellent citation discipline and an API, but its product is search and the API meters per tool. Assistants are strong on execution and weak on retrieval neutrality: ChatGPT and Claude both search, and both are tied to first-party models. Model APIs are strong on the model layer and thin on the rest: you get tokens, not a research surface or an agent rail. Suites are strong on delivery and weak on model choice: enterprise bundles often standardise one vendor.

Riven is designed across all four layers. The model layer is seven Riven-owned lanes served on Riven-owned inference. The retrieval layer is the Research lane, with sources reported per answer. The execution layer is Control and Forge operators, code, agents and the Relay browser under an audited Propose-Approve-Act-Audit rail. The delivery layer is an OpenAI-compatible API at a flat $3.00 per 1M tokens and a self-hosted single-tenant deployment.

Why one platform beats four subscriptions

The economic argument is straightforward. A research seat plus an assistant seat plus an API contract plus an agent tool is roughly $40 to $60 per user per month before API usage, and each vendor has its own admin, invoice and data terms. Riven Pro is $15 per month, Max is $200, teams are $25 per seat, and the API is metered at one flat rate. The saving compounds on the operational side: one identity, one audit trail, one bill and one data-processing agreement instead of four.

The capability argument is that the jobs feed each other. A support answer needs research, drafting and an action; a sales workflow needs research, a document and a CRM update. When those sit on one account with one audit rail, the workflow can be automated end to end rather than handed between tools.

Where point tools still win

A dedicated search engine will usually beat a general platform on pure retrieval transparency, and a dedicated coding assistant will beat a general platform on IDE-specific polish. A single-vendor assistant will beat a multi-lane platform on the depth of its first-party model features, such as image generation or a specific memory design.

The platform argument is not that every component is best in class. It is that the bundle is better than the sum of the four subscriptions most teams currently buy, and that the audit, identity and billing consolidation is worth more than the marginal quality loss on any one job.

What to check before you buy a platform

Ask five questions of any candidate. Can it route per task, or are you locked to one model? Does research cite sources, or is search a tool the assistant may or may not call? Do agents act under an approval and audit rail? Is the API priced per token or per token plus tools? Can it be self-hosted if the data requirement changes? Riven answers yes, yes, yes, flat per token, and yes. Most of the market answers no to at least one.

Point tools vs an AI platform

CapabilitySearch engineAssistantModel APIRiven platform
Model choiceOne engineFirst-party onlyRouted models, no UISeven Riven lanes with Auto routing
Search and researchStrongest citation UISearch as a toolBring your own retrievalResearch lane with sources per answer
Agents and codeBrowser agent onlyAgents on higher tiersBuild it yourselfControl and Forge operators, code, Relay browser
Audit of actionsLimitedLimitedYour responsibilityPropose-Approve-Act-Audit rail
API pricingPer token plus toolsPer model, long-context surchargePer modelFlat $3.00 per 1M tokens, no tool fees
Self-hostedNot publishedNot publishedSome open weightsYes - single tenant
Entry priceUS$17/mo displayedUS$20/mo PlusUsage-based$15/mo Pro ($12.50 annual)

Riven published plan pricing

PlanMonthlyAnnual (per month)What it includes
Free$0$0Seven Riven lanes, starter token grant, chat, code, agents and browse
Pro$15$12.50Full lane access, 1,500 Computer credits, no per-tool fees, 12-month price lock on annual
Max 5x$100$83.335x Pro usage, 5,000 Computer credits, priority routing
Max$200$166.6720x Pro usage, 20,000 Computer credits, highest throughput tier
AI Workspace Team$25 / seat$20.83 / seatShared team workspace, one bill, no seat minimum
Max Team$325$270.8313,000 agentic steps plus $50 API credit per seat per month
API (pay-as-you-go)$3.00 / 1M tokensn/aOpenAI-compatible endpoint, no minimum, add funds from $1

Frequently asked questions

What is an AI platform?

An AI platform is one account that covers several AI jobs - models, retrieval, agents, code and an API - with shared identity, billing and audit, rather than a single-purpose tool. Riven is built as that layer: seven model lanes, a citing Research lane, audited operators, code and an OpenAI-compatible API with self-hosted deployment.

Is Riven an AI search engine?

Riven includes research with citations through its Research lane, but it is a platform rather than a search-only product. The difference is that the same account also carries chat, reasoning, code, agents and an API, so a research answer can be turned into an action without changing tools.

How much does an AI platform cost?

Riven starts free, Pro is $15 per month or $12.50 billed annually, Max is $200, teams are $25 per seat and the API is a flat $3.00 per 1M tokens. Buying the equivalent jobs as separate research, assistant and API subscriptions typically costs $40 to $60 per user per month before API usage.

Can I self-host the Riven platform?

Yes. Riven offers self-hosted, single-tenant deployment on your infrastructure or ours. That is the option the major search engines and first-party assistants do not publish.

Which is better for developers, an API or a platform?

A bare model API is cheaper if you only need tokens and will build your own retrieval, agents and UI. A platform is better when you want the same capability through a UI and an API with one bill. Riven exposes an OpenAI-compatible endpoint at a flat $3.00 per 1M tokens, so you can start with the API and adopt the rest later.

Does an AI platform replace Perplexity or ChatGPT?

It can replace the subscription stack for teams whose work spans research, writing, code and agents. It does not replace the very best dedicated search experience for pure research transparency, and it does not replace first-party image generation. Most teams consolidate what they can and keep one specialist where it clearly wins.

Sources

Related reading