AI writing · checked 2026-10-19Can I vibecode Anyword?
KINDA· weekend project
Generating copy is the easy half: a prompt per channel, a saved brand voice and a few variants take an agent an hour with any LLM key. The half Anyword charges for is the predicted performance score, which it says draws on its own dataset of tested copy and your published results. A personal build has no such data, so the best it can offer is an LLM's opinion of each version, which isn't a prediction. Build the writer, and run real tests where the score would have been.
$49/moprice
$588you'd save a year
one sittingbuild time
—people replaced it
What you need on hand.
DifficultyBeginner · No setup beyond your agent.
TimeOne sitting
- Claude Code, Codex or CursorAny of them can run the prompt.
- Node.js 22better-sqlite3 compiles a native module, so build tools must be installed.
- An LLM API keyAny OpenAI-compatible endpoint works, including a local model.
- Past copy you are proud ofThree examples per brand voice make the output sound like you.
Before you buildAnyword has no free plan; its 7-day trial is capped at 2,500 words. A general chatbot with a saved brand-voice prompt covers much of the writing part.
- 1. Install one agent: Claude Code, Codex, Cursor.
- 2. Make an empty folder and open it in the agent.
- 3. Paste the prompt below, press enter, and answer its questions as it builds.
The prompt
Build me a personal ad-copy workbench with brand voices to replace Anyword. Requirements: - Node.js 22 + TypeScript: Fastify with htmx pages on http://127.0.0.1:4930; storage in better-sqlite3 at ./data/copy.db. - Brand voices: create several, each with a name, audience, tone words, banned words and up to three pasted example texts; every generation prompt includes the chosen voice. - Briefs: a form with product, offer and channel (Google search ad, Meta ad, email subject, landing headline, LinkedIn post), with character limits per channel. The model returns 8 variants as JSON checked with zod; reject any variant over the limit and retry once. - Model: the OpenAI SDK pointed at OPENAI_BASE_URL with OPENAI_API_KEY and MODEL in .env, so any OpenAI-compatible endpoint or a local server works. - Ranking: a second call scores each variant 1-10 for clarity, specificity and fit to the voice with a one-line reason. Label the number "model opinion", never "prediction". - Real results: log which variant ran and its clicks or conversions (typed in or pasted CSV); show winners per voice and channel over time, and put the three best past performers into the next prompt as examples. - Library: save, tag and search variants (SQLite FTS5); export to CSV and Markdown. - No accounts, no telemetry, local-first: only the model call leaves the machine, and the README says exactly what is sent. - Out of scope: a trained performance predictor, publishing to ad platforms, a browser extension, team seats. - README: the .env keys, pointing it at a local model, and a plain note that the scores are an LLM's opinion and only your own results show what works.
Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.
Then keep going
Next prompts, same folder.
The first prompt builds the core. Run these one at a time in the same folder to grow it into your own Anyword.
Add bulk generation to the ad-copy workbench in this folder. Requirements: - Upload a CSV with one product per row; map columns to product, offer and channel, and choose a brand voice for the batch. - Process rows one at a time with at most 20 model calls a minute, store each result as it arrives and let me resume a stopped batch. - Show a progress bar, the number of rows that failed the character limit, and a button that retries only those rows. - Export the whole batch to CSV with one column per variant. Update the README.
Add a results importer to the ad-copy workbench in this folder. Requirements: - Import a CSV exported from Google Ads or Meta Ads: map the columns for ad text, impressions, clicks and conversions in a screen I can save as a preset. - Match rows to saved variants by exact text first, then by a normalized match (lowercase, no punctuation); list the rows that matched nothing so I can link them by hand. - Compute click-through and conversion rates per variant and show them next to the model opinion, so I can see how often the two agree. - Update the README with where to export each CSV.
Add a rule-based lint to the ad-copy workbench in this folder. Requirements: - A rules file (YAML) that I can edit, with checks for ALL CAPS words, repeated punctuation, banned phrases, missing required disclaimers per channel and over-length text. - Run the lint on every variant when it is generated and show warnings next to it; warnings never block saving. - A page listing the most frequent warnings across the library. - Do not claim to match any platform's actual policy; the README must say the rules are mine.
Add a "draft voice from URLs" feature to the ad-copy workbench in this folder. Requirements: - Take up to 5 URLs, fetch them with undici (respecting robots.txt), extract the main text with @mozilla/readability and jsdom, and ask the model for a voice summary: tone words, sentence length, words to use and avoid, and 3 example sentences quoted from the pages. - Show the result in the voice form for me to edit; never save it without my click. - Keep the fetched text in memory only and delete it after the model call. - Update the README with what is sent to the model.
What you lose
- The predicted performance score trained on Anyword's own set of tested copy
- Benchmarks against your past copy and talking-point detection on Business
- Automated website message testing and CMS or CRM publishing
- Shared workspaces with seats, roles and one team brand voice
- A Chrome extension that writes inside other tools
Why people still pay
People pay for the score more than the words: a number that says which headline or subject line is likely to win saves a round of testing, inside a tool that already holds the brand voice. For someone writing dozens of variants a week that can be worth the price; for occasional use, a general chatbot with a saved voice prompt does much of the job.
Anyword or your own copy?
Over
Your usage
3h ×$/h
Anyword, 1 yr$588
Your own copy, 1 yr$65running + build session
You keep$523pays for itself in the first month
Estimate · Runs on your machine. LLM tokens are the only running cost: small for a few sets of variants a week, nothing with a local model. The build is token usage on your agent plan.
Take your data, then cancel.
1 · Take your data out of Anyword
- 1We found no bulk export of documents, projects or brand voices in the help center pages we could read.
- 2Anyword turned older projects into Documents, so copy the text out of each document you want to keep.
- 3Write down your brand voice settings, templates and any saved prompts by hand or with screenshots.
- 4Content already pushed to a CMS, CRM or ad platform stays there.
2 · Cancel Anyword
- 1The support center lists a Billing & Subscription article called How Can I Cancel a Subscription?, but its page returned not found when we checked.
- 2Monthly and yearly subscriptions can be cancelled at any time according to the pricing FAQ.
- 3If you can't find the cancel option in the app, write to info@anyword.com, the address the pricing FAQ gives for subscription questions.
- The pricing FAQ says there are no refunds for cancelled subscriptions, so cancel before the renewal date.
- Starter and Data-Driven both have a 7-day trial capped at 2,500 words; Business and Enterprise go through sales.
- Performance data you uploaded for Business-level models isn't described as exportable: ask support before leaving.
The escape hatch
Don't feel like building it? These are already free.
LibreChatOpen sourceMIT chat interface for many model providers with saved prompts and presets; a place to keep a brand voice prompt.★ 46k · active Oct 2026 · docker to self-host
WebVisit ↗AnythingLLMOpen sourceMIT app that chats over your own documents with local or hosted models; load your past campaigns as context.★ 67k · active Oct 2026 · docker to self-host
Web, macOS, Windows, LinuxVisit ↗Open WebUIOpen sourceSelf-hosted chat front end for local and hosted models; uses its own license with a branding clause rather than a standard OSI one.★ 156k · active Oct 2026 · docker to self-host
WebVisit ↗
WebVisit ↗AnythingLLMOpen sourceMIT app that chats over your own documents with local or hosted models; load your past campaigns as context.★ 67k · active Oct 2026 · docker to self-host
Web, macOS, Windows, LinuxVisit ↗Open WebUIOpen sourceSelf-hosted chat front end for local and hosted models; uses its own license with a branding clause rather than a standard OSI one.★ 156k · active Oct 2026 · docker to self-host
WebVisit ↗
The numbers
Anyword pricing
| Plan | Monthly | Annual, per month | What you get |
|---|---|---|---|
| Starter | $49 | $39 | 1 seat, 1 workspace, unlimited words, 1 brand voice, 100+ templates, Blog Wizard with plagiarism checker, Chrome extension, data-driven editor, 50 performance predictions a month on the monthly view. |
| Data-Driven | $99 | $79 | 3 seats included, 5 workspaces, everything in Starter plus real-time performance predictions for manual edits, 100 predictions a month on the monthly view. |
| Business | custom | — | 3 seats, 10 workspaces, custom-built AI models, benchmarking against past copy, automated website message testing, roles and permissions, onboarding, priority support. |
| Enterprise | custom | — | SSO and enterprise security, a private LLM, custom integrations, full API access, security and legal review, dedicated customer success manager. |
- Free tier
- Trial only: 7 days and 2,500 words.
- Billing
- Monthly or yearly on Starter and Data-Driven; yearly saves 20%. Cancel any time, but the pricing FAQ says there are no refunds for cancelled subscriptions. Plan changes can be made at any time and fees adjust automatically.
- Hidden costs
- Words are unlimited but performance predictions are capped by plan, and the page doesn't say whether the cap resets each month. Extra Data-Driven seats cost $59 a month. The page contradicts itself on brand voices and on Business seats, so confirm both before buying.
- Source
- anyword.com
Also on the list
Related apps
No free plan: a 7-day trial on Starter and Data-Driven, capped at 2,500 words.
Kinda. The core is a weekend build with the prompt on this page, but real gaps remain: The predicted performance score trained on Anyword's own set of tested copy, Benchmarks against your past copy and talking-point detection on Business.
About $49 a month on the Starter plan, $588 a year (checked 2026-10-19).
The predicted performance score trained on Anyword's own set of tested copy; Benchmarks against your past copy and talking-point detection on Business; Automated website message testing and CMS or CRM publishing; Shared workspaces with seats, roles and one team brand voice; A Chrome extension that writes inside other tools. If any of those matter to you, keep paying.
Yes: LibreChat (open source), AnythingLLM (open source), Open WebUI (open source).