ToRun offers individual subscription plans — Free, Starter, Pro, Studio, and Studio Max — plus Teams & Enterprise for organizations and a Partner tier for strategic relationships. Each plan includes a monthly allowance of platform-routed AI usage. When you exceed the included allowance, additional usage is metered platform usage billed from your Wallet at ToRun's published rates. Bring-your-own-key (BYOK) routing has its own three-tier billing model that further reduces platform fees. See /pricing for current plans and rates.

BYOK (Bring-your-own-key) lets you route AI calls through your own provider account — OpenAI, Anthropic, Google, xAI, Mistral and others. The provider bills your own account directly for model usage; ToRun charges only a reduced platform-service fee for routing, moderation and audit. Three BYOK tiers exist: KatmanA (full features, full platform fee), KatmanB (reduced fee for your own account), KatmanC (enterprise tier with audit retention and negotiated rates). BYOK does not exempt you from the Acceptable Use Policy.

Yes. Upgrades take effect immediately and Polar (our Merchant of Record) prorates the difference on your next invoice. Downgrades take effect at the end of the current billing period so you keep the features you paid for. Trial cancellations are not pro-rated. If you upgrade and then change your mind within 14 days of the first paid charge, you can request a full refund under our cooling-off policy.

Visit platform.torun.ai and click Sign Up. We support email + password, magic link, and OAuth / social sign-in via Google, Microsoft, and X (formerly Twitter). Email verification is required for paid features. The Free tier is available immediately on sign-up and includes a recurring free usage allowance; top up your wallet balance any time for pay-as-you-go beyond it. You can add BYOK credentials from Settings → BYOK Vault to route through your own provider account.

Yes. In the web app open the profile menu (top-right) → Personal data — or go to platform.torun.ai/gdpr — and click Delete Personal Data. Erasure runs immediately and cannot be undone: your account and all personal content — chats and memory, uploaded and AI-generated images, video, audio and music (including the stored files), forum posts, workflows, and your profile and avatar — are permanently deleted right away. Records we are legally required to keep for tax and accounting (invoices and billing entries) are retained but anonymized: unlinked from you with personal details stripped. You can export a copy first with Request Personal Data on the same page. Signing up again later with the same email creates a brand-new account — no past chats, balances, or saved content carry over.

Business and Enterprise plans include organization features. From Settings → Organization, an Org Owner can invite members by email; each member gets a TeamInvite that expires in 7 days. Roles are scoped per-team: Org Admin, Team Lead, Member, Billing-Only. Org-level permissions follow our explicit ToRun.Organization..* namespace so each role has a clearly defined capability set. To transfer ownership, both parties must confirm; the change is logged in the audit trail.

We route to most major commercial models — OpenAI (GPT family), Anthropic (Claude family), Google (Gemini), xAI (Grok), Mistral, Cohere, DeepSeek, plus several open-weight models served via our compute fund. The selectable list inside the app reflects which models support the capabilities the current Mode requires. New models are added within days of public release where their pricing has been verified — see /changelog. BYOK lets you reach any provider you have an account with, including ones we don't list by default.

Chat is the everyday conversational interface — best for single-session work, exploration, and one-off generation across text, image, video, and audio. Workflow is a custom DAG engine that turns prompts, tools, and logic into reusable, parameterized automations — best when you want to run the same multi-step operation repeatedly, on a schedule, or across your projects. The two compose: a Chat can spawn a Workflow, and a Workflow can hand its results back to a chat. In either mode, every step routes through the same capability-first model selection and writes exactly one billing record.

A capability is an AI ability — text generation, vision, tool use, structured output, code execution, image generation, audio in, audio out, video generation, embeddings, reranking, search, reasoning, batch, fine-tuning, realtime. ToRun routes by capability rather than by model: you declare what your Mode needs (e.g., Chat needs text + vision + tools), and we pick the cheapest model that supports all required capabilities under your subscription tier. When you switch providers, you don't have to redo the routing — your capability requirements stay the same.

ToRun routes across multiple providers including OpenAI, Anthropic (Claude), Google (Gemini), Mistral, and DeepSeek. Rather than locking you to one model, it uses capability-first routing: you pick what you want to do, and the platform selects an eligible model that supports the required capabilities (vision, tools, audio, etc.) and ranks them by current pricing. You can also bring your own provider keys via the BYOK vault.

Instead of choosing a model by brand, you choose the mode and the capabilities your task needs (for example vision or tool use). The platform filters to models that actually support that capability set, then ranks the candidates by live pricing to pick a cost-effective option. This keeps you on a working model for the job and lets the platform adapt as providers add or change models, without you re-wiring anything.

A knowledge base is a corpus you upload that the AI retrieves from on demand (RAG) — it grounds answers in your documents and can cite them. Memory is self-updating context the assistant maintains about you and your ongoing work, so it carries relevant facts forward across conversations. RAG is "look it up in my documents," while memory is "remember what matters about me." You can use both together.

Yes. Deep research is a multi-source, cited capability: it gathers information across sources and returns findings with citations so you can trace each claim back to its origin. This is distinct from a single chat answer — it is designed for thorough, verifiable research output. Each research run is billed like other AI calls, with one billing record per call.

ToRun's translation covers roughly 100 languages for text, and supports realtime live voice translation for speaking across languages. Separately, the product interface itself is localized into 29 languages. Translation can also work together with memory so context carries across a conversation.

ToRun supports fine-tuning jobs with a self-hosted backend by default, meaning your training data and resulting model adapters stay on your own (or the platform's private) storage rather than being shipped to third-party training buckets. Managed external fine-tuning providers are available only as fallbacks when local capacity is saturated or a base model isn't supported locally. This keeps fine-tuning private-by-default.

Every AI call produces exactly one BillingRecord with a frozen pricing snapshot: provider, model, capability, pricing unit (per-million tokens, per image, per audio second, etc.), exchange rate, and tax rate. The cost is deducted from your included subscription allowance first, then from your Wallet, then from a BYOK pass-through if configured. You can audit every cent in Settings → Billing → Usage. Provider price changes do not retroactively affect already-recorded calls — the snapshot at the time of execution is canonical.

Polar Software, Inc. is our Merchant of Record. Subscription invoices arrive at the start of each billing period; Wallet top-up invoices arrive immediately on purchase. The invoice issuer is Polar, which handles VAT, sales tax, and indirect tax for your jurisdiction. You can download all invoices from Settings → Billing → Invoices. Corporate VAT IDs can be added to your profile so they appear on the invoice. For tax queries that require a Merchant-of-Record document (cross-border reverse charge, etc.) contact [email protected].

Yes, subject to our Refund Policy at /legal/refund-policy. Headline rules: 14-day cooling-off on first paid subscription (full refund, no questions); 90-day dispute window for any charge; unused Wallet balance refundable for 90 days. Email [email protected] with the order ID. We aim to respond within 5 business days. Filing a chargeback without contacting us first may suspend your account during investigation.

Media generation is metered per asset, not by token. Each image, video, or music clip you generate creates exactly one billing record with the provider's price for that output, plus ToRun's transparent margin. The cost is shown for the operation before and after generation, so you always see what a given render costs. Heavier outputs (longer video, higher resolution) cost more because the underlying provider charges more.

Realtime voice features — live translation and voice TTS/STT — are metered per second of audio processed, since that is how the underlying realtime models price. Text translation across roughly 100 languages is handled separately from the per-second voice metering. As with everything on the platform, each call produces its own billing record so you can see exactly what each session cost.

Your subscription allowance is the period-bound usage included with your plan. Your wallet is a prepaid balance you top up that covers usage beyond the included allowance (or pay-as-you-go items). Billing draws from the appropriate fund source per call, and each billing record notes whether it was funded by subscription, wallet, BYOK, or another source. When the wallet runs out, metered overage features stop until you top up.

By design, every AI call creates exactly one billing record — including background work like title generation, summarization, and moderation, not just your direct chat messages. Each record carries a full pricing snapshot (the rate applied, fund source, and a per-capability cost breakdown). This makes your usage fully auditable line by line rather than bundled into an opaque monthly figure.

ToRun uses Polar as the Merchant of Record, which means Polar handles VAT and sales tax collection and remittance on purchases, and issues invoices. You receive proper invoices for your subscription and wallet top-ups through this flow. ToRun is not a financial or tax advisor — Polar is responsible for the tax handling on transactions.

Yes — the billing system supports refunds and dispute handling as part of the Polar Merchant-of-Record flow, alongside invoices. Refunds are recorded against the original transaction rather than editing it, so your billing history stays an accurate ledger. If you have a payment dispute, it is tracked through the platform's billing records so the resolution is auditable.

Platform data lives in primary databases (MongoDB) hosted in the EU (Frankfurt) with cold backup in a secondary EU region. Object storage and CDN delivery use Bunny.net (multi-region edge). Provider calls go to the provider's region as documented in their privacy page — for example, OpenAI traffic typically lands in their US infrastructure regardless of where the request originates. We do not migrate your prompts or outputs to any region other than what the routing requires. Enterprise customers can request EU-only provider routing where the provider supports it.

No. ToRun does not train its own models on your prompts or completions. Provider-side training is governed by the provider's terms — OpenAI, Anthropic, Google all offer no-training options that we use by default on Pro and above. If you bring your own key (BYOK), the relationship is between you and the provider; we do not relay your conversations to any third party for training. The only place your prompts may be retained for non-training purposes is moderation review for severe-violation cases, and even those are accessed only by the responsible safety team under audit.

Open the profile menu (top-right) → Personal data (platform.torun.ai/gdpr). Use Request Personal Data to download a machine-readable copy of everything we hold about you, or Delete Personal Data to erase it. Deletion runs immediately: chats and memory, uploaded and generated media (images, video, audio, music) and their stored files, forum activity, workflows, and your profile are permanently removed. Records required by tax and accounting law (invoices, billing) are kept but anonymized — unlinked from you with personal details stripped. KVKK (Turkey) and equivalent national laws are served by the same page. Deletion is immediate and irreversible.

ToRun supports data-subject rights including data export and erasure for GDPR and KVKK compliance. Content is hard-deleted on erasure, while financial and audit records are anonymized and retained where law requires them to be kept. Requests are tracked formally so there's a record of each access, portability, or deletion request and how it was fulfilled.

Yes. ToRun offers a self-hosted deployment with a one-command Docker installer that builds the backend, database migrator, and web app from source — the host only needs Docker. A single nginx edge handles routing, and secrets and certificates are generated automatically during install. On-prem is aimed at teams that need to keep everything inside their own infrastructure.

Yes. Self-hosted ToRun is designed to work with your own local LLMs, so you can keep inference inside your own environment instead of calling external providers. Combined with the BYOK vault for any external keys you do use, this gives self-hosted operators control over where model calls go. Fine-tuning also defaults to a self-hosted backend, keeping model weights on your own storage.

Yes, for organizations that need maximum assurance ToRun offers a source-code license option and a source-code escrow arrangement, in addition to the standard one-command self-hosted deployment. These are aimed at customers (for example regulated institutions) that require continuity guarantees and the ability to inspect or maintain the code. Specific commercial terms are handled through the on-prem licensing process.

BYOK (bring your own key) lets you store your own provider API keys in an encrypted vault and use them for your AI calls. It is organized into three tiers, and ToRun applies a transparent margin in the range of 1.3x to 1.7x on metered usage. Using your own keys changes how calls are funded and priced; the exact tier applied is recorded on each billing record.

Moderation runs through a tier-based router that selects an appropriate provider based on your subscription tier, rather than calling one provider directly from app code. CSAM scanning via PhotoDNA runs at every tier with no bypass — there is no plan level that turns it off. Each moderation check is itself metered and recorded, so moderation activity is transparent and accountable.

Sensitive audit and billing reconciliation records are hash-chained: each entry includes the hash of the previous entry, so the chain can be recomputed to detect tampering. A periodic job verifies the chain and flags any break. This gives financial and security-relevant events tamper-evidence, which matters for compliance and trust in the billing ledger.

The in-app assistant can take actions inside the product on your behalf, and it is built around an approval and undo model: it requests approval for actions and supports undoing them, so it doesn't make irreversible changes silently. Its tool invocations are recorded for auditability. This lets you delegate multi-step tasks while keeping a human in the loop.

Projects organize your work into separate spaces. No-code workflows let you build automations as a DAG (a graph of steps) without writing code — chaining AI calls and tools into a repeatable pipeline. Any AI steps inside a workflow are billed per call like normal usage.