Because the major AI coding tools moved to token-based billing in 2026, so you now pay per token your assistant reads and generates. Agentic and repository-wide work can burn 50,000 to 200,000 tokens in a single task, and heavy users have seen monthly bills go from tens of euros to many hundreds. Pauhu Fusion builds the heavy parts off the token meter, so that share of the bill goes flat.

Yes. Pauhu Fusion is a flat 279 EUR per seat per month, falling the more you scale. You keep your existing assistant for the intent; Pauhu builds the code without spending model tokens, so that work stays outside your variable per-token bill.

The biggest lever is to stop generating, token by token, what a compiler can build. Pauhu Fusion hands your copilot that compiler: it builds code from published standards without a model call, so those calls cost no tokens, and the receipt shows the tokens and energy each call avoided. Your model only handles the small intent part.

On token-based plans it varies with usage: individual developers commonly land between tens and a few hundred euros, heavy agentic users report 200 EUR and well beyond, and enterprise reports reach around 2,000 EUR per engineer per month. Pauhu Fusion is a flat 279 EUR per seat, falling with scale, for the work it does, with the saving shown on the receipt.

Yes. Pauhu Fusion is a remote MCP server your existing assistant calls. Keep Copilot, Cursor, Claude Code, or Codex; add Pauhu with a Bearer key and your assistant calls it for the code it can build. The per-client setup is on For developers.

No, and this is the difference that matters. A probabilistic model can hallucinate a library or function that was never real, the 2026 security risk known as slopsquatting, where attackers register the invented package with malware. Pauhu never invents an API: every line it builds traces back to a real published source, so it cannot suggest something that does not exist.

No. Pauhu is a compiler, not an AI model, so there is nothing to train and nothing learns from your code. Your repository stays in your own tool; only the part a request needs crosses to Pauhu, processed in the EU, and the repo itself never leaves your machine.

Move the heavy, repeatable work off the variable token meter. With usage-based AI coding, spend tracks how hard your team works that month, which is how some teams have burned an annual budget in a quarter. Pauhu Fusion is a flat 279 EUR per seat, falling with scale, for the work it does, so that line stops swinging and the receipt shows what each call avoided.

No. Pauhu Fusion is a Model Context Protocol (MCP) compiler. Your coding assistant reads your repository and decides what a request needs. Pauhu builds code from published standards; the same request returns the same result every time. The only AI in the workflow is your own assistant.

No. Your assistant holds the repository and sends only the part a request needs. The entire repo never crosses to Pauhu. The compiler runs in the EU and returns the code and a receipt.

No. Pauhu works alongside your assistant. It builds the code so your assistant can focus on the intent. Your assistant still performs the natural-language to code step and handles any gaps.

No. It is not a general chatbot or agent. It does not replace a senior engineer, does not browse your files, and cannot run commands. It builds the code and returns it to your assistant.

Context-trimming tools reduce the data sent to an LLM by heuristically selecting relevant files. Pauhu builds the requested operation from published sources. It does not guess or summarise your repository; only the part a request needs crosses the wire.

Pauhu builds code for common operations across files, strings, data, and the web, against published standards and library APIs, and the library keeps growing. It never invents an API: every line traces back to a real source.

Your own assistant handles it exactly as it does today. Pauhu never returns fabricated output, and the receipt shows which calls Pauhu built at no token cost.

No. Pauhu builds code from published sources, but it does not guarantee the code is bug-free, secure, or suitable for your task. You are responsible for reviewing, testing, and validating the output before production use.

The work it does does not call a model or run a GPU. If you then send the code on to a separate LLM of your choice, that provider pricing applies to that part. Pauhu’s receipt records both paths.

Pauhu is a remote MCP server at https://mcp.pauhu.dev. You need an API key, issued once we let you in. On clients like Claude Code, VS Code Copilot, Cursor, or Codex you add the server URL and Bearer token; examples are on For developers. Setup takes a few minutes and needs no plugin install.

Yes. Pauhu sits in front of the AI you already use. You can keep or swap your copilot; the compiler is the same and your repository stays in your tool.

A Model Context Protocol (MCP) server is a standard way for an AI assistant to call an external tool, a universal connector between your copilot and a service. Pauhu Fusion is a remote MCP server at mcp.pauhu.dev: add it once with a Bearer key and your assistant calls it for code it can build from published sources.

Currently Pauhu Fusion is a hosted service at mcp.pauhu.dev and is not available as an on-premise or offline deployment. Your repository stays local, but the work happens on the hosted server. Self-hosting may be considered later but is not offered today.

Yes. The subscription is for ordinary individual or team development use. Fair-use and rate-limit rules are published in the Acceptable Use Policy; sustained automated abuse may lead to throttling or suspension.

As of June 2026 there is no published uptime SLA. Status for the MCP endpoint and website is on the status page linked from Community. Support is by email at hello@pauhu.ai. Paid support or priority SLAs may be introduced later.

A flat 279 EUR per user per month, and the per-seat price falls the more your team grows. Billing is month to month, with no annual lock, the European way: Peppol e-invoice, SEPA. We are setting up payment and invoicing now; reserve your place and we will let you in the moment it opens.

We are setting up payment and invoicing. Reserve your place from the pricing page and we will let you in the moment it opens, billed the European way: Peppol e-invoice, SEPA.

Cancel anytime via the billing portal; access continues until the end of the paid period. Because the subscription is month to month, partial months are not normally refunded; charges made in error will be refunded. Consumers in the EU/EEA may have a 14-day withdrawal right.

The service is licensed per user at a flat monthly price. For larger teams or enterprises needing purchase orders, Stripe Invoicing supports B2B and B2G invoicing at the same price. Volume discounts or private deployments are not currently advertised; contact us for custom arrangements.

You own the code Pauhu returns and may use, modify, and distribute it subject to the licences of any sources it draws on. Pauhu retains all rights in its corpus and engine. Provenance markers may be removed for production use, but you must comply with upstream licences.

Yes. Pauhu AI Ltd is the data controller and processes account data, request payloads, usage receipts, and truncated IP addresses as described in the Privacy Policy, under contract and legitimate-interest legal bases. Your repo does not leave your tool. The work runs in the EU; the only sub-processors are Hetzner (hosting, Finland) and Stripe (payments, USA). Exercise your GDPR rights by emailing dpo@pauhu.ai.

The compiler is a tool that builds code and does not learn or make automated decisions about users. The imprint states it is not an AI system within Regulation (EU) 2024/1689. Because it does not train or adapt, it does not fall under high-risk AI obligations. Any AI in your workflow is your own assistant under its provider terms.

The compiler runs in Hetzner’s Finnish data centre. Requests are transmitted over TLS. Only the part a request needs is sent; your repository stays in your tool. API keys can be rotated and revoked. The DPA lists sub-processors and locations. Pauhu does not currently publish SOC 2 or ISO 27001 attestations but follows GDPR and implements technical and organisational measures.

No. The site uses no external or third-party scripts. Cookie usage is limited to strictly necessary storage and the payment provider. There are no advertising or analytics trackers.