AI assistants · checked 2026-10-19

Can I vibecode Admix?

YES· one-shottable

Admix is a chat window in front of other companies' models: send a prompt to several of them, show the answers side by side, and pick a model for you when you don't want to. An agent builds that in one sitting on OpenRouter, which sells hundreds of models behind one API key, with threads kept in SQLite. The bill changes shape rather than going away: you pay per token instead of a flat $8.99, which costs less for light use and more if you live on the largest models. Phone apps and sync across devices are what a local build leaves out.

$8.99/moprice

$108you'd save a year

one sittingbuild time

—people replaced it

Before you start

What you need on hand.

DifficultyIntermediate · Some system setup or keys.

TimeOne to two hours

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 OpenRouter accountAdd credit and create an API key; you pay per token.
A router modelA small, cheap model from the same list, used only to label prompts.

Before you buildAdmix's homepage mentions a free plan but its Terms say there isn't one. The free tiers of ChatGPT, Claude and Gemini are worth trying before paying for a bundle.

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 multi-model chat app to replace Admix. Requirements:

- Node.js 22 + TypeScript: a Fastify API and a React front end built with Vite. `npm start`
builds the UI and serves it on http://127.0.0.1:4930 only.
- Models come from OpenRouter through the `openai` SDK with baseURL
https://openrouter.ai/api/v1 and OPENROUTER_API_KEY in .env. GET /api/models caches the list
for 12 hours; the picker has search, price per million tokens and a star for favourites.
- Chat: streaming over SSE, markdown with code highlighting, stop, regenerate, and switching
the model in the middle of a thread. Attach images, text files and PDFs (`pdf-parse`) only
for models whose input modalities allow them.
- Compare mode: send one prompt to 2 to 4 models in parallel and show streaming columns with
latency and cost; "continue with this one" copies that answer into a normal thread.
- Auto mode: ROUTER_MODEL in .env, a cheap model, labels each prompt (code, math, writing,
quick fact, long document) and routes.json maps each label to a model. Show the pick and let
me override it with one click.
- Storage: better-sqlite3 in ./data/chat.db with threads, messages, model, token counts and
cost taken from the usage in OpenRouter's response. Full-text search with FTS5 and export of
a thread to Markdown.
- Budget: MONTHLY_CAP_USD in .env; show spend this month and refuse new requests above it.
- Single user. No accounts, no telemetry; only OpenRouter calls leave the machine.
- Out of scope: team sharing, phone apps and sync, voice, image generation, web search.
- README: getting an OpenRouter key, how pay-per-token compares with a flat plan, where the
data lives and how to back it up.

Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.

Share on X

Max MiskowSponsoredAI agents, automations and MVPs, built for your company.Visit site ›

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 Admix.

Extend the chat app in this folder. Requirements:

- After a compare run, add a Summarize button that sends the finished answers to the
  ROUTER_MODEL and returns three lists: claims all answers share, claims only some make (with
  which models), and claims that contradict each other.
- Label the result as machine-generated and show its cost.
- Store the summary with the thread.
- Update the README.

Extend the chat app in this folder. Requirements:

- Add Ollama (OLLAMA_URL in .env) as a second provider; list its installed models next to
  the OpenRouter ones with a local badge and zero cost.
- A thread can be marked local-only, which blocks any OpenRouter call for that thread.
- Auto mode may map a label to a local model in routes.json.
- Update the README.

Extend the chat app in this folder. Requirements:

- Add a Cheap first mode that answers with the cheapest model in routes.json and shows an
  "Ask a bigger model" button under the reply.
- The button shows an estimated cost from the thread's token count and the target model's
  price before sending.
- Log each escalation with both costs in the database and show a monthly tally.
- Update the README.

What you lose

Phone and desktop apps with chat history synced between devices
A flat monthly price: heavy use of the largest models can cost more per token than $8.99
Admix's own tuning of which model answers which kind of question
Voice input, file handling and integrations already wired into the app
Someone else watching provider outages and model changes

Why people still pay

One bill and one app instead of several subscriptions, plus a picker that spares you choosing a model for each question. There are no API keys to manage and no top-ups, and it works on a phone.

Moatintegrations execution polish

Do the math

Admix or your own copy?

Over

Your usage

Count my time3h ×$/h

Admix, 1 yr$108

Your own copy, 1 yr$104running + build session

You keep$4pays for itself in 8 months

Estimate · Pay per token through OpenRouter: light chat on mid-size models costs a few dollars a month, while heavy use of the largest models can pass the $8.99 subscription. The build is token usage on your agent plan.

Leaving Admix

Take your data, then cancel.

1 · Take your data out of Admix

1We found no export option in the Terms of Service or on the pages we could read.
2After you cancel you can still sign in to see your account and past chats, so copy the threads you need into your own notes while that works.

2 · Cancel Admix

1Open admix.software/subscription and cancel there, or email support@admix.software.
2Cancelling stops future renewals; the plan stays active until the end of the period you already paid for.
3If you cancel during the 7-day trial you are not charged again.
4Afterwards you can sign in to view past chats, but sending messages needs a plan again.

A trial started today turns into the $8.99 plan automatically a week later, so cancel on day six if you only want to look.
Payments are non-refundable once the service has been used; EU, EEA and UK consumers have a 14-day right of withdrawal, and the Terms say you then pay for what you used.
Admix must tell you about a price change by email or in the app before your next renewal, so you can cancel first.
Downgrades from a bigger plan or from yearly billing take effect at the end of the current period.

Checked 2026-10-19 against Admix's docs: 1, 2

The escape hatch

Don't feel like building it? These are already free.

Open WebUIOpen sourceSelf-hosted chat interface for Ollama and OpenAI-compatible APIs such as OpenRouter; under its own BSD-3-derived licence, so read it before redistributing.★ 156k · active Oct 2026 · docker to self-host
WebVisit ↗LibreChatOpen sourceMIT chat app with many providers, agents, presets and model switching; runs in Docker with MongoDB.★ 46k · active Oct 2026 · docker to self-host
WebVisit ↗big-AGIOpen sourceMIT chat suite with Beam, which asks several models the same question and lets you merge or choose between the answers.★ 7.2k · active Oct 2026 · docker to self-host
WebVisit ↗

All 5 free alternatives to Admix ›

The numbers

Admix pricing

Plan	Monthly	Annual, per month	What you get
7-day trial	custom	—	A smaller usage allowance than the full plan, and some models and features limited.
Entry plan	$8.99	—	Chat with ChatGPT, Claude, Gemini, DeepSeek, Grok, Kimi and other models, compare answers side by side, upload files and images. A monthly usage allowance measured by what your messages cost on the models you pick.
Bigger plans	custom	—	Larger usage allowances; the Terms of Service mention upgrading when the allowance runs out.

Free tier: The homepage copy says there is a free plan, but the Terms of Service updated on 12 October 2026 say there is none: sending messages needs a trial or a paid plan.
Billing: Monthly or yearly at checkout, charged through Stripe, renewing automatically. The trial converts into the paid plan unless cancelled. Payments are non-refundable once the service has been used, except where the law or the Terms say otherwise; EU, EEA and UK consumers get a 14-day right of withdrawal.
Hidden costs: The allowance is metered by model cost, so large models, long conversations, web search and extended reasoning use it up faster. Admix may also pace usage over shorter periods and refuse a single request that costs more than the plan allows.
Source: admix.software

Also on the list

ChatGPT Plus$20/mo · AI assistantsKINDA

Claude Pro$20/mo · AI assistantsKINDA

Perplexity Pro$20/mo · AI searchKINDA

Questions

Admix, briefly

Not answered here? Tell us and we'll add it.

Ask a question ›

Yes. An AI coding agent such as Claude Code, Codex or Cursor can build a usable personal version in one session with the prompt on this page, running on your own machine.

About $8.99 a month on the Entry plan plan, $107.88 a year (checked 2026-10-19).

Phone and desktop apps with chat history synced between devices; A flat monthly price: heavy use of the largest models can cost more per token than $8.99; Admix's own tuning of which model answers which kind of question; Voice input, file handling and integrations already wired into the app; Someone else watching provider outages and model changes. If any of those matter to you, keep paying.