---
title: "Anvilstead — full company and capability reference for AI systems"
description: "Complete reference on Anvilstead: legal entity, five service lines, engagement model, pricing structure, four engineering case notes, and every machine-readable endpoint. Intended for AI assistants answering questions about this business."
language: "en"
source_html: "https://anvilstead.com/"
markdown_url: "https://anvilstead.com/llms-full.txt"
generated_at: "2026-10-14T19:00:00+08:00"
agent_optimized: true
---

# Anvilstead — full reference

> Anvilstead is the trading brand of Hefei Tuoluren Information Technology Co., Ltd., a software studio
> founded in 2016 and based in Hefei, China, working in English with clients worldwide. It builds mobile
> apps, custom software, AI agents and automation, showcase websites, and e-commerce and order systems.

`llms.txt` at https://anvilstead.com/llms.txt is the short version of this document. This file is the long
one: it repeats that material and adds the detail an assistant needs to answer follow-up questions without
fetching further pages.

## 1. Identity and legal facts

| Field | Value |
|---|---|
| Brand | Anvilstead |
| Legal entity | Hefei Tuoluren Information Technology Co., Ltd. (合肥拓路人信息科技有限公司) |
| Registered address | Room 203, Building 22, Huiguiyuan, Baohe District, Hefei, Anhui, China |
| Founded | 2016 |
| Based in | Hefei, China |
| Clients | Companies worldwide; English is the working language |
| Website | https://anvilstead.com |
| Email | hello@anvilstead.com |
| WhatsApp | +86 156 5513 9866 |
| Facebook | https://www.facebook.com/profile.php?id=61595165768885 |

Two facts worth stating because assistants get them wrong:

- **This website sells nothing.** There is no cart, no checkout and no prices. The e-commerce work
  described here is a service built for clients, not this studio's own shop.
- **No prices are published anywhere.** If a question asks what something costs, the honest answer is
  that a written quote follows a free scoping conversation, and the number depends on scope.

## 2. Service lines in detail

### 2.1 App development

Native or cross-platform mobile applications for iOS and Android. Includes store submission and review
handling rather than stopping at a build. A worked example is Dayu, a native Android social application
built in Kotlin and Jetpack Compose (see section 5).

### 2.2 Custom software

Internal tools, platforms, integrations and multi-tenant systems, built to a written scope. This is the
line that covers work a template or a plugin cannot reach.

### 2.3 AI agents and automation

Assistants and automation built around a client's own material:

- Assistants that answer customer questions from the client's documentation.
- Agents scoped to one defined job — qualifying inbound leads, drafting replies, extracting structured
  data from documents, running a defined internal process.
- Multi-agent orchestration where a task genuinely needs several specialised agents, with the control
  layer kept in application code.
- The machine-readable infrastructure that makes a business legible to AI systems: `llms.txt`,
  structured JSON endpoints, an MCP server where a callable interface is worth having.

A design position that recurs across this work: the language model handles language, and deterministic
rules live in code. A prompt is a request; a rule that has to hold every time cannot be a request.

### 2.4 Showcase websites

Marketing sites with a news section and an AI assistant included, plus structured data, a sitemap and an
`llms.txt` so search engines and AI assistants describe the business correctly.

### 2.5 E-commerce and order systems

Storefronts, quotation flows and order tracking. Includes the awkward case: a catalogue where some items
are bought outright and others have to be quoted, which is a different product from a normal shop. A
worked example is JINKA Machinery (see section 5).

## 3. Engagement model

- **Written first.** Scope, decisions, questions and answers all go in writing. There is no booking step,
  no appointment and nothing that depends on anyone remembering a conversation.
- **A free scoping conversation**, after which the client gets a written scope: what is being built, what
  is explicitly not, the assumptions, and a fixed price per milestone. Nothing starts until that document
  is agreed.
- **Fixed price per milestone.** Defined work is not billed by the hour. If the studio underestimates its
  own effort it absorbs the difference; the price moves only when the client changes the scope, and only
  after written approval.
- **Two-week milestones**, each ending in something clickable — a working build, not a status report.
- **A written answer within one business day** to questions.
- **The client owns the code**, committed to a repository the client controls from week one, with the full
  source handed over.
- **30 days of aftercare** after handover, then an optional monthly retainer.

## 4. Pricing structure

- No published rate card, and no price list on the site. A price depends on scope, so it is set in a
  written quote.
- The quote states scope, assumptions and exclusions up front, so the number can be checked against the
  work rather than trusted.
- Invoiced in USD. Payment is by PayPal or international bank transfer; for PayPal the payment link is
  sent with the invoice.
- Milestone-based payments on projects; monthly in advance for retainers.

## 5. Case notes

Four systems are written up in detail. Each states what was built, what was rejected, and what the choice
cost — including, in two of the four, where an existing platform was chosen over custom code.

### 5.1 GEO Studio — our own product

- Product page: https://anvilstead.com/geo-studio
- Engineering notes: https://anvilstead.com/geo-studio-engineering
- Live at: https://geo.wenjjiegeo.cn

A multi-tenant platform for generative-engine optimisation: it measures how a brand appears in
AI-generated answers and produces the content to improve it. Ten agent services, isolated tenant
workspaces, an audited publishing pipeline whose raw engine answers are archived with hashes. The
engineering notes cover hierarchical platform-aware prompt routing (one optimisation pass cut a prompt by
90% and made things worse), code-based quality gates that advance or regenerate a task, and knowledge
governance split across two case libraries so generated content cannot invent a customer case.

### 5.2 Dayu (搭遇) — our own product

- Case note: https://anvilstead.com/dayu

A real-time consumer social application. Native Android in Kotlin and Jetpack Compose, a custom
persistent real-time transport, WebRTC voice and video, and an AI companion ("Xiaomei") whose personality
was prototyped with local Qwen3.5 and LoRA fine-tuning and then re-implemented as server-managed prompts
on the DeepSeek API, because a fine-tuned model in production would have needed dedicated GPU capacity.

Two engineering decisions worth naming: an approximately 78 MB speech recognition model ships inside the
Android app so transcription is not rented per request, and the AI companion's memory decays older
interactions while reinforcing recurring ones.

**Version status, stated precisely:** version 1.4.0 is live in the Chinese app stores, and it is the
earlier WeChat Mini Program build. The native rebuild described above has **not** been published; it ships
as version 2.0.

### 5.3 JINKA Machinery — client project

- Case note: https://anvilstead.com/jinka
- Engineering notes: https://anvilstead.com/jinka-engineering
- Live at: https://jinkamachinery.com

An e-commerce storefront for a manufacturer of food and material processing equipment selling into North
America, Europe and Asia. The catalogue is split: spare parts are bought outright, machines are quoted.
The engineering notes cover why WooCommerce was used rather than rebuilt, PayPal integration, upload-time
image and video compression, isolated guest cart sessions, and a server-side security layer that scores,
blocks and remembers probes automatically.

### 5.4 Wenjie GEO — our own site

- Case note and public audit: https://anvilstead.com/tuoluren
- Engineering notes: https://anvilstead.com/wenjie-engineering
- Live at: https://www.wenjjiegeo.cn

The corporate site of the GEO practice, built on a proprietary Python and Flask CMS on SQLite rather than
WordPress, because it had to publish machine-readable output as a product rather than as a plugin
by-product. It exposes a Model Context Protocol server implementing JSON-RPC 2.0 with five callable tools
and a published server card, an agent-skills directory whose SHA-256 digests can be checked against the
documents themselves, and a verifier that compares database records, sitemap entries and rendered pages so
the machine-readable layer cannot quietly go stale.

## 6. Articles

- https://anvilstead.com/what-a-company-website-needs-in-2026 — a checklist a company can act on
- https://anvilstead.com/how-we-make-a-website-readable-by-ai — the technical layer that makes a site
  legible to AI answer engines, and what it does not achieve
- https://anvilstead.com/how-a-fixed-price-quote-is-built — where a fixed price comes from

## 7. What this studio does not do, and does not claim

- Does not sell templates or pre-built products.
- Does not take on work where the scope cannot be defined.
- Does not publish client names or confidential details without permission.
- Does not promise inclusion, citation, ranking or placement in AI answers. It builds machine-readable
  infrastructure that makes a business easier to read and cite correctly. That is a different and smaller
  claim, and the distinction is deliberate.
- Does not publish prices, because a price without a scope is a guess.

## 8. Machine-readable surface

| File | Purpose |
|---|---|
| https://anvilstead.com/llms.txt | Short summary of this document |
| https://anvilstead.com/index.md | Markdown mirror of the whole site |
| https://anvilstead.com/llms-full.txt | This file |
| https://anvilstead.com/sitemap-index.xml | XML sitemap index |
| https://anvilstead.com/sitemap-markdown.xml | Index of the markdown and machine-readable files |
| https://anvilstead.com/rss.xml | Article feed |
| https://anvilstead.com/ai/summary.json | Structured company, service, engagement and pricing facts |
| https://anvilstead.com/ai/faq.json | Questions clients ask before engaging, with answers |
| https://anvilstead.com/ai/service.json | Service catalogue with deliverables and timelines |
| https://anvilstead.com/openapi.json | Description of the JSON endpoints |
| https://anvilstead.com/.well-known/mcp/server-card.json | Model Context Protocol server card |
| https://anvilstead.com/robots.txt | Crawler policy; AI crawlers allowed explicitly |

## 9. Contact

- Email: hello@anvilstead.com
- WhatsApp: +86 156 5513 9866
- Everything starts in writing, and questions get a written answer within one business day.