A new version of Ask Archie is available.
Chat is locked
0 here right now
Set a code to lock this chat on this device.
Closed for now — check back soon.
A brain before the brief.
Talk it through with Archie first, work out what you actually need, then get it done without wasting credits along the way.
The AI can change, but the project stays locked in with Archie.
Get started ↓Illustrative figures — not yet final. See real, current pricing below.
Real, current figures — not illustrative.
Everything in one purchase: local coding unlock, a hosting slot, the local-AI onboarding add-on, and £10 starting credit. For someone who already knows they want it all from day one — not a discount incentive, same total as buying each part separately unless a promotion is running.
already has an Ask Archie account. Enter its existing password to link your Google sign-in to it — this is a one-time security check, never done automatically just because the email matches.
No spam — just updates as Ask Archie develops.
Real, live output from whatever Archie Companion is currently running — polled every few seconds, never a fabricated "still working" placeholder.
Shown before it ever runs on your machine — nothing executes here without your explicit approval, every time, no exceptions.
This is separate from your AI credit — it unlocks local coding on your own machine, using your choice of coding agent and Archie's own free local guide model where your hardware can run it. Every other part of Ask Archie is already available to you, same as anyone else — this only unlocks the local-coding capability specifically.
£10, one-time, non-refundable, lifetime. This is not spend credit — it never runs out and never touches your brief-building balance. Marketed at this rate for a limited time; the price may rise later.
Archie: Hi — I'm Archie. Let's get your account set up, then get your first project started. No rush.
The AI can change, but the project stays locked in with Archie.
Gemma is a small local AI that checks your machine and helps Archie understand what it can handle, before you start.
Download Gemma — a checksum-verified file from Ask Archie's own servers.
Once downloaded, run it locally to get her working, then come back here.
Skipping this will limit what Ask Archie can offer you — without her, some capabilities won't be available. You can always install her later, and you can uninstall her at any time. She never runs silently, and never does more than the one simple job she's here for.
Checking for a local Gemma helper...
Once her check completes, this step shows the real result — what she found about your machine, plainly, not just a "done" message. The same result also appears on Gemma's own page in Settings, always kept in sync with what's shown here.
This is the display name other people on Ask Archie will see for you — never your email.
Want to join the Community too — post, join groups, share your work? That's optional, and you can verify your identity any time from your Profile once you're in.
Skip for now →This step matters most if you're planning to do local coding on your own equipment. Tell Archie your country and supplier so he can use the best available regional pricing when estimating local compute costs — and suggest cheaper times of day for your heaviest work, where that region's pricing changes through the day. A direct connection to your own personal tariff (Octopus, Tibber, etc.) is planned, not built yet — for now, estimates are based on your country and supplier's typical regional rate.
Pick your country and supplier below. Skip for now and set this up any time later from Settings — it's never a one-time, locked-in choice.
Skip for now →Choose the reasoning model you want to work with. You can change it later.
Archie isn't the model. He coordinates your project, keeps its context together, and works with whichever model you choose.
The AI can change, but the project stays locked in with Archie.
Skip for now →Already have your own key from a provider you trust — Anthropic, or any of the others we support? Add it here, or skip and add it later from Settings.
No API key is ever shown in plain text, anywhere, at any point — masked on entry, masked in any later display, genuinely deleted (not soft-hidden) if removed. Same standing requirement as everywhere else on the site a key appears.
Skip for now →If you're here for local coding, this is your step. Run a coding agent right on your own machine.
You can let a smaller reasoning model guide your code through a coding agent — cheaper than using a premium model for every step, and free if you're running your own local AI already. Or, if you'd rather not choose, let Archie recommend the best fit — he'll weigh your project against available models, cost, and your machine's capability. You still decide; Archie never picks for you.
Slow burn option, mentioned here too: if you're not in a rush, slow burn mode runs your build at the lowest possible priority on your machine — it takes longer, but never gets in the way of anything else you're doing at the same time.
Not doing local coding right now? Skip this and set up a coding agent any time later from Settings.
Skip for now →If you ever want to build your own audio plugin with VST Builder, you'll need your own GitHub account connected -- that's where the real source lives and the real Windows/Mac build runs. Not needed for anything else on the site, so skip this if it's not for you.
Already have a GitHub account? Generate a personal access token at github.com/settings/tokens/new (tick repo and workflow) and paste it below.
By now you should have a good sense of how Archie works. Time to create your first project — we recommend starting with something simple, just to get a feel for the platform before you dive into anything bigger.
Give your project a name and a short description. You can paste your ideas into its Waiting Room right after, and add more as you go.
Real, existing open-source scaffolds — not built from scratch. Locked templates remain visible, so you can see what's available before unlocking them.
Every project you own — in progress, parked, published, or sold — in one place. A parked project shows its storage location right on its own row.
Sell your apps in the App Store, or transfer with confidence through Escrow.
Your hosting slots — a free one comes bundled with the £10 local-coding unlock; add more at £5/month each, standalone or as a second (or third) project.
Your app is live now — give it a name, and watch it go live at that exact address.
Required before buying a domain — you become the legal registrant of record, not Ask Archie. Collected once, reused automatically at purchase time.
Paste a link — a spec doc, a reference site, anything with content on it. Archie reads the page and proposes a Final Brief inspired by it, the same as if you'd pasted it into chat yourself. You still review and accept before anything gets built.
This will delete:
This will not be deleted (stays visible, independent of this project):
This cannot be undone.
Confirming again — deleting will delete:
This cannot be undone. There is no further confirmation after this.
Every note, generated image, completed video, and uploaded audio file you own, in one place — search it, tag it, tidy it up, and send an item straight to a project or a group feed without hunting for where it lives.
Nothing here yet.
Whole Synth module snapshots saved from the Synth screen -- all three voices, both drum units, Master, Song, Cross-patch.
No saved sessions yet -- save one from the Synth screen's own header.
Every prompt you've sent across Ask Archie -- Chat, App Builder, Compute, Let's Study, Image Maker, Video, Synth, Music Lab, and app amendments -- kept as a searchable reference back to what you typed and where it took you.
No prompts logged yet -- use Chat, Image Maker, Music Lab, or any other Archie feature and they'll start showing up here.
Browse apps other users have listed. Log in to contact a seller, fork a free app, or buy.
No listings yet.
Off by default. When set, the app locks itself behind your password after this many minutes of inactivity -- worth turning on if you keep spendable credit on this account and use a shared or unattended device. Your session stays signed in; this only re-checks that it's still you.
Defaults to the "Ask Archie" paint-splash image -- or pick any image from your own Vault instead.
Instead of one still image, slowly crossfade through a set of your own Vault images with gentle drifting movement.
Selected for the slideshow (pick at least 2):
Top up from £10, paid securely through Stripe — or choose a package to get more from Ask Archie. Keep track of your balance throughout the site — you'll always know where you stand.
Real, live connection status for each — never a static list. Connect or disconnect any of these at any time.
Live credit balance funded by your top-up — used automatically whenever a generation isn't covered by one of your own connected provider keys below.
Powered by Stripe
A real, per-build spend ceiling for Deep Build (Archie's coding agent) -- once a build's real cost reaches this, it finishes its current step and stops cleanly rather than continuing to spend. Applies separately from your overall balance above.
A one-time government-ID check via Stripe Identity — required before buying a domain, and before any Community feature (feed, chat, groups, briefs, App Store sharing).
Route your calls through your own Anthropic/Google/Mistral/DeepSeek/Qwen key instead of the shared pool, using our BYOK mechanism. Free to connect — spend for these calls goes straight to your own provider account, not Ask Archie's balance above. If your own provider balance ever runs out mid-conversation, Archie will let you know honestly and explain what topping up would actually get you.
Your own personal access token, generated in your own GitHub account — never Ask Archie's own GitHub, which is admin/site-only. Used to push a project's export when you choose "List it on GitHub," and required for VST Builder — every plugin project lives in a real repo of your own, since that's where the actual build runs.
Your own Anthropic API key, from your own Anthropic account — never a shared or default Ask Archie credential. Used only to authenticate your own Claude Code CLI runs inside your own Sandbox session. Without this connected, cloud coding execution simply can't run for your account — there is no fallback.
Your own Cloudflare API token and account ID — an alternative to Ask Archie's own Extended Hosting (£5/month): your build deploys into your own Cloudflare account instead, which you already pay for separately, so there's no Archie-side charge for this destination.
Stripe Connect account, in your own name — required before listing anything for a price (a one-off sale or per-download). Ask Archie takes a commission per sale; the rest pays out straight to your own Stripe account, never held in your Ask Archie balance. Free listings don't need this.
A persistent default link, set once — used automatically for new parked projects unless you set a different link on that specific project.
Future enhancement, not built yet — would provide your exact personal electricity rate instead of the regional Elecz estimate. Flagged for later, not blocking anything here.
Off by default for everyone. Turn this on to see raw diagnostic detail on execution failures, instead of the simplified view.
Invite-code pairing and an approval-gated command queue, executed by a local Companion process you run on your own machine — approved commands run there and report results back. Folder scope is enforced, not just recorded: a project-scoped device can only run commands inside the folder you consented to. Deliberately kept out of sign-up entirely — you opt into this here, on your own terms, once you actually want it.
If you're an individual: only pair devices you personally own and control. An invite code grants access to approve commands against your own projects — treat it like a password, not something to share.
If you're using this for a business or team: Ask Archie has no multi-seat account system yet (see Settings → Team, not yet built) — every paired device is authorized under your individual account with your individual permissions. There's no per-employee scoping or audit separation between team members at this stage; don't pair a shared/company device expecting individual accountability.
On the machine you're pairing, open a terminal (PowerShell on Windows, Terminal on Mac/Linux) in your Companion folder and run:
node companion.js
Paste your invite code when it prompts for one. Your browser should open automatically to confirm — if it doesn't, copy the URL it prints instead.
Once pairing succeeds, the Companion badge in the top nav lights up green and reads "Ready". If you ever restart your computer, that badge will show "Paired, not running" (pale orange outline) until you run node companion.js again — your pairing itself is still valid, the process just needs to be started.
To stop it: press Ctrl+C in that terminal window, or click "Stop device" below (list further down this page) to stop it remotely from here. Either way your pairing stays valid — running node companion.js again resumes it, no new invite code needed. "Revoke" below is different: it permanently ends the pairing and requires a fresh invite code to reconnect.
Kept separate from infrastructure connections above — these are people, not services. Manage requests and messaging from the Community tab in any project workspace; shown here for a current overview.
Exactly what Ask Archie's browser-ping check found on your machine — the same data used internally, never a separate hidden dataset.
Every real charge type this account can have, pulled live — nothing here is cached or estimated.
A real, one-time payment — never a recurring subscription. Provisions or tops up your own model-provider key with a spend cap matching your payment, real and enforced by the provider itself, not tracked-only.
Small usage charges can build up between top-ups rather than being collected right away. Opt in to let Ask Archie collect what you owe from a saved card once it crosses a small minimum — a deliberate, one-time consent, never automatic unless you say so. You can withdraw this at any time.
Optional — connects to a local AI runtime on your own machine.
A guided Q&A session with Archie, scoped only to the real Knowledge Base — good for learning the platform properly.
Which model powers Archie's proposal generation for you — a choice from the available pool of models.
Not limited to the one model you started with — browse whatever's actually loaded in your own local runtime (LM Studio or Ollama) right now, and choose a different one for local offloading. A real, live catalog from your machine, never a fixed list.
Want the guided setup — browsing models, install links, and a live hello-world check once loaded? That's a one-time £5 add-on.
An OpenAI-compatible /v1/chat/completions endpoint — use your own scripts/tools against the same models Ask Archie itself uses. Draws on your own provisioned AI provider balance (top up first if you haven't), same cost as everywhere else in the app. Rate limit: 20 requests/minute per key.
Copy this now — it won't be shown again.
Add up to 5 real links to your other pages — YouTube, SoundCloud, Spotify, Facebook, Reddit, and more. Shown as a real icon row on your profile, so people can find you elsewhere too.
Findable by display name in Community search by default, so other developers on the platform can find you. Turn this off if you'd rather not be found — you can still search for and message others either way.
Your location or supplier can change — come back and update this any time, not just at sign-up.
Live pricing for your set location — the same red/green cheap-window signal Archie uses when factoring cost into your build.
Your balance is shown in the top bar on every page by default. Turn it off if you'd rather it stayed private -- your balance is unaffected either way, this only controls whether it's shown.
Locked after inactivity
Pick one — both give you a local server Ask Archie can talk to once a model's loaded.
Get LM Studio → (graphical, easiest to start with)
Get Ollama → (terminal-based, lighter)
A real, live search against Hugging Face's own catalog — never a fixed or fabricated list. Look for "GGUF" in the name/tags for LM Studio/Ollama compatibility.
If you're already running your own local AI, Ask Archie works around your current GPU load instead of assuming it has the machine to itself.
Not reported yet.
Once it's loaded and the runtime's local server is running, this checks for it the same way Settings does — a genuine /v1/models request, never assumed.
Sends one small message to your local model and shows its reply — proof that it's working, not merely detected.
"Hello, I'm Gemma — your bounded local helper. I was installed on your workstation in a lightweight configuration to help establish your work environment and connect you to the network. You can uninstall me at any time. Leaving me installed means I'm available for future updates, health checks, and workspace checks without having to set me up again."
Gemma 3 1B (270M as a fallback for very light hardware) is downloaded directly from Ask Archie's own servers — a specific, tested, verified copy, not something you have to find or gate-check yourself elsewhere:
1. Install LM Studio (recommended for this first step — graphical, no terminal). Ollama works too, if you already have it.
2. Download Gemma 3 1B directly from Ask Archie — a real, checksum-verified file, never a proxy or a guess.
3. Load the downloaded file into LM Studio (or Ollama) as a local model, then start its local server and come back here to check below.
Redistribution notice · Gemma Terms of Use (as provided to you, per Section 3.1)
Not checked yet.
Exactly what the browser-ping capability check found on this machine — the same data used internally, here too, not just in Settings.
GPU data — memory, power draw, utilisation — read directly from your own machine via nvidia-smi. Ask Archie's cloud backend has no OS process access, so this can't be triggered from a button here: run node scripts/gpu-status.js --post <your session cookie> from a terminal on your own machine (see the repo's scripts folder). Genuinely optional — nothing here blocks anything else.
Not reported yet.
Discovering compute across other authorised machines on a network — business or academic/research authorised-network scoped only (e.g. a university department pooling lab machines), not active for individual accounts. Needs its own security/legal review before going live.
Verification unlocks Community participation, including posting, joining groups, and sharing your work.
Stripe Identity handles the document check directly — Ask Archie only ever sees whether it succeeded, not your documents.
✓ Verified
Add a photo of yourself and a cover photo so people recognise you — takes a minute, and you can always change it later.
Photos you've shared into Community — pulled from your Vault or uploaded fresh, right here. Hover a photo to remove it.
Nothing here yet.
No matching images found.
Every video edit you've rendered lands here automatically — no upload step. Share up to 2 a day to Community.
Nothing rendered yet — finish an export in the Video editor and it'll show up here.
Synth sessions and Music Lab tracks you've sent to your profile — unlimited, private to you unless you share the link.
Nothing here yet — use "Send to Profile" from the Synth screen or Music Lab.
No friends yet — search above to connect with someone.
One group per person, to keep moderation simple — you can still join as many others as you like.
Nothing posted yet — share something above.
Shown to your friends so they know when it's coming up — the year is never shown or shared, just the day.
No upcoming birthdays from your friends yet.
Nothing posted yet.
Full chat, image upload, and brief-sharing tools live in Community — this feed is the shared, read-first view of what's shared there.
Nothing here yet — shared images, briefs, and App Store listings will show up here as one unified feed.
You're not in any groups yet — create your own on your profile, or search below to discover one to join.
No pending requests.
No matching images.
No matching videos.
No posts yet.
Admin-only reference notes — distinct from the public-facing Knowledge Base.
When enabled, login and sign-up are disabled for everyone except admins.
Narrower than the lockdown above: existing accounts and logins keep working normally. Only new public sign-up is paused — a test code is the one way to create a genuine test account while this is on. While this is on, admin/super_admin accounts can also reach local execution in the handover pipeline, so an end-to-end walkthrough is possible without exposing it to real users.
Infrastructure only — no recording pipeline is built yet. Starting a call/broadcast while this is on logs a "recording requested but no pipeline built" line server-side, with no other side effects, so the switch can be tested safely before the actual pipeline exists.
Real client-reported detections only — never a server-side claim about what's installed on anyone's machine. Network discovery: gated, not active for individual users.
A real, changeable link to the actual R2 bucket in your Cloudflare account — not a synthetic in-app view standing in for it.
Open R2 dashboard →Ask Archie's own R2-hosted copy of Gemma — a deliberate pin to one specific uploaded revision, never an auto-following "latest" link. Reversed away from direct Hugging Face linking after account-gating friction was found there; the file is hosted and served from Ask Archie's own R2 storage. Updating the pin here is a logged admin action.
Open the R2 download link →R2 multipart upload — handles the full ~769MB file. Nothing goes live until every part uploads, R2 confirms the stored size matches, and the pin updates automatically.
Which services each user has connected — status only, never the credentials themselves.
Infrastructure names, kept out of user-facing copy by design — visible here for admin/ops reference only.
Real revenue, real cost, real profit — built from data already tracked. Nothing here is fabricated or estimated beyond what's honestly labeled below.
User-submitted reports on Community content — images, briefs, groups, and messages. Resolve once you've dealt with it; dismiss if it isn't one.
No pending reports.
The one credit pool behind everything — every per-user provisioned key and the house/admin pool key both draw down against this same balance. Checked automatically every 30 minutes; top-up itself always stays a deliberate manual action on OpenRouter's own site.
Top up on OpenRouter →Every package that draws on genuine Archie coordination gets its own, independently-adjustable rate — never one shared number applied everywhere by coincidence. Each fires on every OpenRouter call made through a provisioned key (chat, voice, polish passes, image generation, hosting deploys, advisory checks, cloud handover) — which package's rate applies depends on whether the account is coding-only-tier-unlocked (Charge 2) or running on a standard top-up alone (Charge 1).
No rate control here, deliberately — hosting is a flat monthly fee with no per-use event to apply a trickle-style rate to today. Adding a numeric field with nothing behind it would be decorative, not a working control.
An adjustable minimum — no per-user override (below, on each customer's card) can go below this, protecting against a well-meaning discount that accidentally becomes a loss-making rate.
Site-wide rate changes above never touch existing manual per-user overrides — they're fully independent. Use this if you deliberately want every manual discount reverted back to the site default, e.g. after a rebrand/repricing.
One checkout bundling local coding unlock (£10), a hosting slot, the local-AI add-on (£5), and £10 starting credit. No discount by default — set a live-calculated discount below if wanted.
Price for the full local-AI onboarding page (model browser, guided setup, hello-world check) -- only purchasable by accounts that already have the £10 coding-only unlock.
Amounts owed. A row only shows as charged once an off-session collection has succeeded (below) — never assumed.
Deliberately manual for now — each charge below is admin-triggered against a specific account's consent and owed total, not an unattended sweep. Only accounts that have opted in (Settings) and crossed the minimum threshold are chargeable.
Current source docs behind the Hosting marker brief — check directly against current Cloudflare state rather than a point-in-time summary.
Future unification (hosting + balance/billing + Companion handshake tokens into one coherent Accounts area) is logged, not being built now — see incoming-briefs/STATUS.md.
The £4/month standalone Extended Hosting tier has been retired — replaced by the £10 bundle (includes one free slot) and the £5/month standalone slot below.
An on-demand check each time this loads (not a persisted Cron history like the OpenRouter balance monitor).
A daily snapshot (Cron), plus this per-prefix breakdown from R2 directly.
Never overwrites the current version — every publish creates a new, permanently dated row. History below.
Never overwrites the current version — every publish creates a new, permanently dated row. History below.
Select a user on the left to see their full record.
Consolidates every App Store admin function in one place — content moderation, stale-listing lifecycle, storage overview. Auction monitoring, tender-assignment, and buyer-protection groundwork are described in the spec but not built yet (no bidding/tender/protection mechanism exists to administer) — listed here honestly rather than left silently missing.
Checkable conditions only: the underlying project was deleted but the listing wasn't removed, or the seller hasn't logged in for 90+ days. Never removed automatically — your explicit choice each time.
Genuine current storage consumption (DB content + R2 objects), not a flat estimate — the actual figure the "Cloudflare storage" line in any future profit view should draw from.
NOT IMPLEMENTED — none of these mechanisms exist yet in the build (no bidding, no tender/checkpoint-assignment, no buyer-protection flow). Build rules stated honestly rather than shown as empty working sections.
Every card on the front page — location, content, add, edit, remove. Nothing here is hardcoded.
The real, explicit allowlist of facts Image Maker can use when "Use site context" is on (default on) — plus the same public Knowledge Base Chat's own toggle uses. Nothing else reaches the image model. Only real facts belong here, never a guess.
Structured, editable sections.
The articles shown on the public /knowledge-base page — editable here, no code deploy needed for a correction.
Defined page locations — a slot only ever shows where it's actually positioned for.
Fast, direct navigation to the external tools used to run the business.
Admin-only — served through the admin API, never in the public bundle. Nothing here is public-facing until deliberately made so.
Ongoing tracking — sits alongside sign-in/acquisition since both drive growth. Persistent status, never a one-off checklist re-run from scratch.
Checking sitemap.xml/robots.txt status...
Not automatable here (no Search Console/LLMrefs API access) — check manually: Search Console (Core Web Vitals, indexing, crawl errors) · PageSpeed Insights · LLMrefs GEO Score.
Send a notification, site-wide or to one specific user — informational, never engagement-optimised (spec §14).
Bulk send via Resend to Subscribers, Partial sign-ups, or both — deduped by email, and anyone who's unsubscribed or already converted to an account is automatically excluded. Every email includes a working unsubscribe link.
Email capture from both visitors and signed-up users — genuine interest, not a fabricated list.
Email captured the moment it's entered on the sign-up form, even if the flow is abandoned — re-engagement candidates. "Send follow-up" sends an email via Resend directly to the captured address (there's no account here, so nothing can be shown in-app) and logs it as contacted automatically. For sending to everyone at once, use Email campaigns above instead.
Configurable slots — hidden entirely on the public site when no sponsor is assigned, never a visible empty placeholder.
The honest baseline: UTM parameters captured at sign-up. For deeper reporting, use GA4 and Meta Ads Manager's own dashboards externally — not built here, per the founder's own right-sizing call.
Measured usage of each BYOK provider across the user base — evidence for a funding or partnership conversation, not a cold pitch.
NOT IMPLEMENTED — stated honestly rather than left silently missing: print document/image upload tooling, camera/microphone hardware-access pathways, a social-share module generator, provider logo rows (needs a Brandfetch integration, deferred from the User API Keys brief), and a founder-only marketing tab linked to Notes. None of these have a backing mechanism in this build yet.
One unified view: the same tier grouping shown to users, plus fleet-wide install/usage stats from every Companion check-in — new admin context, not duplicated from the per-user Models page.
Chat always has it. These switches decide whether it also shows on the other pages that offer a model choice — each one steps that page's own model selection up by one, from whatever's already chosen.
Accounts whose repeated, unresolved Stripe Identity verification attempts crossed the flag threshold — a pattern noticed and tracked, not buried in generic error logs. Stays listed even if the account later verified successfully.
The gap-fill continuity mechanism and the formalized handover surface registry — call-based, cost only when actually invoked. No live memory service, no vector index running.
Every model and agent option in one place — a real, browsable pool, not a fixed default.
Your app is live — open it or .
Below, you'll see your brief develop with every accepted response from Archie. In the Archie chat, you'll see Accept or Reject on each reply — click Accept and it appears below as a Mini-brief. Want to remove one? Untick its Mini-brief card over on the bottom left and it comes back out of the document. ⓘ
Not resolved automatically — you decide how to proceed with each.
Your resume package is your way to carry this conversation forward — hand it to a new model if you switch mid-build, or take your brief with you out of Ask Archie into your own workspace. Either way, nothing gets lost.
Per-attempt results from the verify-before-done check — did it load and click through clean, how many attempts it took, and the actual error text on a failure. Not a substitute for trying the app yourself, just the same evidence Archie used to decide whether to call it "ready."
Narrows within whatever scope is set above (this project, or all your notes) — file type and date filters, combined with the search above.
Image generation — genuinely more than chat, gated behind Charge 1. Auto-saved to your Vault either way; tie it to this project to make it a build asset.
Once listed, it shows up in "Browse listings" below for anyone to find — yours and everyone else's live in the same place.
Choose a model, or if you have free credit you can use them first.
If you decide to choose a new model during your build, please don't forget to generate a resume package — you'll find that over on your right at the bottom of the page.
Out of date — regenerate firstYou've been away — want me to catch up before we continue?
Waiting Room — empty
This includes an image — Archie will use a vision-capable model to look at it instead of your selected model. Cost shown afterward as usual.
Notes, images, and videos you've deliberately added to this project — feeds the coding agent's own context when it builds, separate from the brief conversation above.
No assets added yet — add one from your Vault.
Seller-provided — not independently verified by Ask Archie.
Use Ask Archie's own Escrow.com integration (payment held until the seller genuinely delivers), meet or deal with the seller directly, or use your own separate escrow service. None of these is the default.
Your app as it's actually deployed, re-checked every time you open this tab and after every build.
Nothing built yet — it'll show up here the moment a Deep build finishes.
Open in a new tab → Your app is updating…
⚠ Separate AI cost — shown once it finishes.
Free — no AI call.
Develops here with every accepted exchange.
You've got at least one accepted mini-brief but no Final Brief yet -- Deep build needs one before it can run.
Your Final Brief is ready -- nothing's been built yet.
A short, honest walkthrough of every stage in this chat -- extend this as we go.
Type in the box below (or paste in notes/ideas) and send. Archie turns what you say into a proposed exchange -- a concrete addition to your brief, not a decision made for you. Accept it and it becomes a "mini-brief"; reject it and nothing changes. The more exchanges you accept, the more complete your brief gets -- there's no limit, and no rush. Add as many as you genuinely need.
Once you've accepted at least one mini-brief, a "Compile & mark Final Brief" step reads everything you've accepted so far into one coherent Final Brief -- the actual document Deep Build works from. This is a cheap pass (not the same as Polish, below), and it's required before a first build can run at all.
A separate, genuinely optional pass that re-reads your whole accepted brief and reconciles any contradictions (e.g. you changed your mind on a colour halfway through) into one clean, coherent version. It's a separate AI cost every time you run it, shown up front before you commit. Worth it for a large or genuinely conflicting brief; skip it for a small, simple one -- Update app and Build this now never trigger it automatically, so you're never charged for a polish pass you didn't ask for.
Build this now appears once your Final Brief is ready but nothing's been built yet -- it kicks off the very first Deep Build. Update app takes over once your app is genuinely live -- it reruns the same Deep Build agent against your current Final Brief, seeded from what's actually deployed, so it extends what's there rather than starting over. Neither ever spends on a Polish pass by itself -- if your Final Brief is stale when you click Update app, it just does an honest, zero-cost join of everything accepted since, no LLM call involved.
Any message you send once your app is live is treated as a targeted change request -- Archie's Amend agent reads your app's actual current files first, makes a minimal edit (not a full rewrite), and re-checks the live result afterward. It tells you plainly how many tool calls it made and what it actually changed.
Tested live: a genuinely detailed, upfront brief (an 11-voice drum machine, full spec written out in one message) produced a strong, working first build in one pass -- the core structure, look, and feel all landed correctly. But the more repetitive, granular parts of that brief (per-voice controls, multiple pattern banks) got quietly trimmed rather than flagged as skipped, and needed separate amendments afterward to actually land.
The lesson: get the build right at the beginning. Write the first brief to establish a solid foundation -- what the app fundamentally is, its core structure, and its overall look and feel -- rather than trying to list every last detail. Once that foundation is live, bring in the finer details and extras one at a time as amendments. Each one lands more reliably on its own than as one more line buried in a long list, and you can actually see and approve each change as it happens.
Next to Polish sits a "Try a bigger model?" button. First click gives you a cost estimate, free -- second click actually runs it. It re-reads your whole accepted brief (every mini-brief, not just the last one) through a genuinely more expensive model, and gives you a much more thorough, detailed brief document in return.
Tested live: it did produce a noticeably richer brief -- more than double the length of the regular Polish pass. But when that brief was actually built, the result was almost the same app, with one small fix -- because the staged, one-at-a-time amendments that got us there had already covered the substance, for a fraction of the cost. That's the honest finding, not a guess: a well-staged brief, built up through ordinary free/cheap exchanges, gets you most of the way there on its own -- a bigger model refines the wording, it doesn't replace the groundwork.
So: build your brief in good, deliberate stages first (see above) -- that's free, and it's what actually determines how little "fuss" (follow-up amendments) you'll need later. Save "Try a bigger model" for genuinely large or tangled briefs where you want a second, more careful read -- not as a substitute for staging it well in the first place.
Let's Study has a "How to use Archie" preset -- a Socratic Q&A session scoped only to the real Knowledge Base, if you'd rather learn this by working through it with Archie than by reading. Open Let's Study →
More stages will be added here as the build flow grows -- flag anything missing or unclear.
Setup for how this build actually runs — pick your coding agent, execution, and model below, or hand off to your own machine or cloud agent further down.
You can run your build on your own machine, with your own AI coding assistant, instead of through us. We'll hand you the brief and a secure way to connect. But once it's on your machine, it's your machine, your network, your call — and your responsibility. Same honesty applies to cloud execution: what leaves the Vault, what the coding agent can access.
Your coding agent:
There are bots constantly scanning for exposed remote access left badly secured — not aimed at you specifically, just automated and relentless. Roughly 3 in 10 real breaches trace back to weak or reused credentials, not clever hacking. Locked down properly (short-lived token, scoped folder, an agent you actually trust), this is a different thing entirely. If you're a business, get your own IT/security person involved before switching this on.
Cloud offload consent is on record for this project. The eight-part handover script, packaged from this Final Brief, is sent to an LLM here in the cloud for automated code generation -- not literally the CLI tool you picked above, which is what actually runs your coding agent of choice when you offload to your own machine instead. Execution/verification stays a separate, honest, evidence-required step below — nothing here is ever assumed working just because it generated.
Cloud generation builds directly from your Final Brief above — nothing to re-describe.
Eight-part handover script, packaged from this Final Brief. Once queued, your paired device picks it up — every command it runs (including the build itself) is shown in the Live Build Viewer for your explicit approval first, no exceptions.
Your chosen coding agent (above) builds directly from your Final Brief above — nothing to re-describe.
Stopping the sandbox never risks your work — checkpoints let you roll back to any prior state regardless of what happens here.
A record of every command actually run in the sandbox — not a live build viewer, since no automated build-execution pipeline runs commands here yet.
Free static preview shows what it looks like. A real hosting slot (bundled free with local coding, or £5/month standalone — manage in Settings) is the real thing running. Every deploy is checked for Workers-runtime compatibility first.
This project has no hosting slot assigned yet. Assign one from Settings → Hosting, or connect your own Cloudflare account for a free deploy.
A decision point — fallback to cloud (with cost shown first) or stop dead and review — never a forced or silent switch. Shared across two genuinely different triggers: execution failure (crash/out-of-memory/stalling) and environment integrity problems (malware/disk-full/workplace-disabled).
Sends the Final Brief's text directly, via Community messaging — not a literal drag-and-drop (no such mechanism exists elsewhere in the app yet to reuse), but the same messaging channel.
Couldn't set up your chat space just now — try refreshing.
Choose which model Archie chats with you as. You can switch any time — nothing about your conversation is lost when you do.
A guided study session with Archie — Socratic questions, not straight answers, so you build your own understanding instead of copying his.
Something went wrong setting up your study space — please try reloading the page.
Our default starter model is great for topic introductions and getting your bearings quickly. For deeper academic insight, switch to a larger, more complex model above.
Studying:
Save this understanding — saves your own answer in your own words, then Archie asks the next question.
Move on — skip this question with no answer given; Archie asks a different one instead.
Let's discuss — get a straight explanation for just this one question, then try answering it yourself.
Compile study guide — once you're done, turns everything you've saved into one recap document.
The Companion is the local bridge between Ask Archie and your own machine -- running code on your own hardware, with your own coding agent. Two separate pieces get you there:
Come back here — Companion will take you straight into your project's Live Dev View, ready to pair your device and start local execution.
Costed and tracked the same as any other generation — spend against your provisioned key, no simulated results.
Your most recent generation will show here.
Every generation saves to your general Vault first — nothing is attached to any project automatically. Use "Send to Archie" on a generation below to put it in a specific project's Waiting Room, whenever you actually want that.
Every note and generated image across every project, plus anything general — the same search a project's Notes tab uses, reachable from here too.
Drag on the image to select the area to keep.
Saves as a brand new image in your gallery — the original is never changed.
Each generation is an 8-second scene by default — combine several and build it further with our editor below. Generate longer scenes using the duration input selector below.
Combine your generated scenes, voiceovers, your own music, and your own video — matched automatically to your project's resolution. Click your project link below to go to the editor.
Thorough (Sonnet 5) is what we suggest -- it's more likely to follow every instruction in one request: the song structure, the automation/movement, and the tone/filter shaping, all at once. It costs more per request and runs slower.
Fast (Haiku) is perfectly fine for a single, simple edit -- a tone tweak, one pattern change, a quick adjustment. But if you're asking for more -- a multi-part arrangement, automation, tone shaping -- give it to the more experienced assistant instead.
This is a per-request choice, not a one-way setting -- switch back to Fast any time.
Loads this clip's current (trimmed) audio onto the chosen drum voice -- replaces whatever sample is loaded there now, same as uploading a sample directly on that voice.
Log in to ask Archie to shape or write something -- the keyboard and controller above are yours to play freely either way.
Play or upload a reference sound -- Archie's tone choices get grounded in what's actually measured from it, not a guess from the words alone.
Upload a real .mid file -- import a chosen bar straight onto a voice's steps, or ask Archie above to build around it.
Record, upload, or generate (via text-to-speech) a real vocal clip -- it plays alongside the composition (Play/Play Song), with its own Volume/Mute/Solo in the Mixer below.
Archie's real reply -- what it just set up, or the tone change it made -- shows up here.
One real channel strip per unit -- Volume/Pan/Mute/Solo at a glance. Each strip's own controls are the exact same real ones as that unit's own section below (not a separate copy) -- jump straight there for its full effects chain.
A 303-style step sequencer -- 16 steps, per-step note/octave/accent/slide. Plays through the exact same filter, filter envelope, rack, and master chain as everything else above -- no separate sound path. Slide gives a genuine glide (a smooth pitch ramp between notes, one sustained voice, no retrigger), the defining acid-line technique.
The final stage -- everything above passes through here before it reaches your speakers or a recording. Trim controls level into the limiter below (fast, catches overs so stacking unison/resonance/distortion can't clip); the meter reads the actual post-limiter signal live. Recording taps from this exact point -- true output, not a re-capture of individual parts.
Reorderable signal chain -- top to bottom is genuinely the audio path, move a module and the sound actually reflects it. Start basic (bypass everything), enable and reorder as you go deeper.
Route one unit's trigger to modulate another, like a virtual patch cable. Every time the Source unit plays a step, the Destination unit's own level briefly ducks and recovers. Set Amount to 0 to turn it off.
Chain each unit's own saved patterns into a song -- pick a pattern (or "Hold" to keep what's already loaded) per unit for each step, and how many bars to stay there before moving on. Play song restarts the transport from bar 1 so the timing lines up.
Play it yourself too -- click a key, or use it while Archie's composition plays.
Real audio nodes in the signal path, not a display gimmick -- every note played by hand, MIDI, or previewed by Archie goes through this. Voice A and Voice B each have their own separate filter above -- not shared with this one.
Save the whole current setup -- oscillator, unison, filter, rack, master -- and search back to it later. Saved per-browser, same as your controller mapping.
Connect your own USB/MIDI keyboard -- real, live input, not a plugin. Every controller maps its knobs differently, so instead of guessing at one brand's layout, wiggle a knob and use Learn to bind it yourself; it works the same way for any controller.
Quantized -- what you play snaps to the current step, straight into the sequencer. Loop stays on to overdub across passes; only the steps you actually play get written.
A separate, experimental page for trying AI music generation against a prompt -- Archie's own engine, or an external music model. Not yet a settled feature.
Try Music Lab →Upload, record, or generate a real vocal phrase -- it lands on the timeline below as a clip you can drag into position, nudge, trim, and split. Effects and automation land next. Shares the same transport and tempo as your Synth composition.
The full Synth module -- three real 303-style voices, drum units, mixer, effects, and song mode. Synth is where you can build a track; Music Lab is for a quick, real generated clip.
Describe a track and get real, generated audio back -- a choice of model, audio out. Costs money against your provisioned key.
Log in to generate real audio -- this is real spend against your own provisioned key.
Every clip you've generated with Lyria, saved to your Vault automatically.
Every VST project needs a real GitHub repo of your own -- that's where the actual C++ source lives, and where the real Windows/Mac build runs. Ask Archie never touches your account beyond what you connect here.
repo and workflow scopesChoose a name for your VST. ⓘ
Pick the layout your new plugin's own interface should use -- do this before pressing "Create a VST project" below, since it's set at creation time.
This shapes how Archie thinks about your request from the start.
Archie only ever selects and parameterizes real, pre-tested JUCE DSP primitives from this library -- never freeform-generated C++.
Tone
Effects
Click a scene below to add it to your timeline.
Click an image below to add it to your timeline -- 3 seconds by default, adjustable per-clip. From your Image Maker gallery -- never locked to one project, add the same image anywhere you like.
Add your text at the playhead and move the playhead to set the end point. Use the controls below to colour and position.
Drag a transition onto the boundary between two clips on the timeline.
Fine-tune duration, or set without dragging:
Fade and speed, per clip.
The video clips' own soundtrack, baked in from each clip -- not a separate uploaded file. Plays live during preview. Use the header's volume/mute/solo to adjust its level.
No music added yet.
Drag the waveform's edges to trim. Trims to your video's length either way.
Click a clip below to use it as this track -- from your Vault (uploads, Synth sessions, Music Lab tracks), never locked to one project.
No voiceover added yet.
Type the words and get a real, ready-to-use voiceover -- no mic or upload needed. Real spend, same as any other generation.
Drag the waveform's edges to trim. Trims to your video's length either way.
Click a clip below to use it as this track -- from your Vault (uploads, Synth sessions, Music Lab tracks), never locked to one project.
The canvas every clip gets fitted to. Set automatically from your first clip; change it here any time.
Renders entirely in this browser tab — keep it open until it finishes.
Scenes staged here, not yet on your timeline. Organise, add to the timeline, or share when you're ready.
Saved to your general Vault — tagged with this project so you know where a note came from later.
No notes yet.
Open My Vault →The real, live picture — nothing here is cached beyond what's noted.
Loading...
You can switch this back on any time in Settings.