Most AI tools have a core tension they never acknowledge: the best model for your task today may not be the same model tomorrow, yet you are expected to commit to one provider's subscription and work around its blind spots. The billing is opaque. The tools are scattered. When something goes wrong, you cannot tell whether you were charged fairly.

ToRun is built to fix that.

The problem we are actually solving

Single-provider lock-in is not just a pricing inconvenience. It means your chat assistant, your image generator, your code tool, and your document pipeline each live in a different tab with a different billing statement and a different mental model. Switching providers when a better model ships means migrating workflows and re-learning interfaces. And throughout all of this, the typical platform gives you a monthly dollar figure with no way to trace which call cost what or whether the rate was accurate on that date.

These are not minor UX complaints. They compound into real cost and real risk for anyone using AI at scale.

How ToRun works

ToRun routes every request through a four-stage pipeline: the product Mode you are in (Chat, Code, Image, Research, and others) determines the Capability set required (text generation, vision, tool use, structured output, image generation, audio, and more). That capability set filters the full model catalog — which spans OpenAI, Anthropic/Claude, Google/Gemini, DeepSeek, xAI/Grok, Mistral, and others — down to only the models that can actually do the job. From that filtered set, a price-and-quality ranking picks the model, and a fallback chain takes over automatically if the chosen provider is unavailable.

You do not pick a model. You describe what you need. The platform finds the one that delivers it at the best trade-off of cost and capability.

Transparent billing is not a marketing claim here — it is a hard architectural constraint. Every single AI call writes exactly one billing record with a frozen price snapshot captured at execution time. If a provider changes its rates six months from now, your historical invoices are still fully verifiable and recomputable offline. Twelve canonical pricing units cover every modality (per million input and output tokens, per image, per second of audio and video, per character of TTS, and more). You can export the ledger and check every cent independently.

BYOK (bring your own key) lets you paste your own provider API keys into an encrypted vault. ToRun resolves them per call without ever exposing them to application code, audits every read, and lets you scope a key to a specific provider, model, capability, or team. Three billing tiers trade platform fees against routing through your own provider account — useful for teams that have negotiated enterprise contracts directly with a provider.

What is live in Phase 1

The platform ships today with:

• Multi-provider chat across all supported modes, with automatic capability routing and fallback chains
• Generative media — image, video, and music generation as first-class modes, billed per image or per second, not via a separate product
• Memory — long-term facts about you persist across sessions, are user-owned, user-controllable, and work across all 29 supported locales; you can list what is remembered and delete any of it
• Knowledge bases — ground answers in your own documents via RAG retrieval
• Workflows — a visual DAG builder for chaining models, tools, and logic into automations; runners select quality-tier models by default
• Marketplace foundations — creators can list workflows with a revenue share starting at 55% for Pro-plan creators
• Personas — AI synthetic creators with subscriptions, tips, and brand-deal infrastructure; mandatory disclosure is enforced before any persona can post or be listed
• BYOK vault with per-key scoping and a full audit log
• Wallet + subscription allowances — prepaid balance draws down per use; free-tier sessions refill through the day; premium features gate on whether you have funding, not on a rigid plan flag

What comes next

Phase 2 — in active development, no dates committed — adds Council (multi-model deliberation with an adjudicator), DeepResearch with per-claim source citations, the full public workflow-marketplace listing flow, and the bouncyad.com ad exchange. These are not shipping today.

If you have spent time managing three separate AI subscriptions, wondering why your bill jumped, or rebuilding the same automation in three different tools, ToRun is the platform that was missing. The architecture is honest by design, not by marketing copy.

See /pricing for current plan details and /docs to get started.