Free conversation, paid artefacts
mode=sketch costs nothing: it answers in your language and names the products it would use. You only pay when you ask for the blueprint or the code.
Tell cn8 Architect what you want to build — "transcribe my videos and let
Learn more about cn8 ArchitectThe conversation so far: [{role: user|assistant, content: string}]. Write in any language; the answer comes back in the same one.
Instant sample preview — no cost, no account. Sign up free to run with your own input.
{
"status": "success",
"data": {
"mode": "blueprint",
"message": "Videolarınız için üç adımlı bir hat kurdum…",
"plan": {
"id": "3f1c…",
"title": "Video Q&A over transcripts",
"nodes": [
{
"id": "t1",
"productKey": "ai-transcript",
"name": "Transcription",
"service": "transcript-transcribe",
"role": "Turn each uploaded video into subtitles.",
"body": {
"video_url": "https://…/lesson.mp4",
"subtitle_formats": [
"vtt"
]
},
"produces": "subtitle_urls.vtt",
"facts": {
"method": "POST",
"path": "/v1/proxy/transcript-transcribe",
"async": true,
"price_label": "0.1 cr/second"
}
}
],
"external": [
{
"id": "store",
"name": "Your object storage",
"kind": "storage",
"role": "Holds the videos and the extracted text."
}
],
"edges": [
{
"from": "t1",
"to": "ds1",
"payload": "VTT subtitle file",
"glue": "Poll the job, download the VTT, strip the timestamps."
}
],
"prerequisites": [
"A Management API Key is required to create data sources."
],
"validation": {
"errors": [],
"repaired": false
}
}
},
"cost": {
"tokens": 50,
"unit_price": 50,
"balance": 4950
}
}Example response from this endpoint's API reference. Sign in to run it on your own input.
This request costs 0.5 per item.
Tell cn8 Architect what you want to build — "transcribe my videos and let people ask questions about them" — and it answers with the architecture: which cn8 services, in which order, what travels between them, and what you still have to build yourself.
It plans against the live catalog, not against a model's memory. Every service name and every request field in a blueprint is checked against the same schemas that power the API reference, and anything that does not match is sent back to the planner before you see it. The code step goes further: the API calls in the generated project are produced by the gateway's own sample generator, so the URL, the auth header, the JSON quoting and the async polling loop are correct by construction — the model only writes the glue between them.
Three modes, one endpoint. sketch is a free conversation that names the products. blueprint is the paid architecture. code turns a blueprint into a runnable project in the language you pick.
mode=sketch costs nothing: it answers in your language and names the products it would use. You only pay when you ask for the blueprint or the code.
Product keys, service names and request fields are checked against the live schemas. Unknown fields are removed, missing required fields are reported, and the planner gets one repair pass before the answer is returned.
Your storage, your database, your frontend, your scheduler — the blueprint models them as external boxes instead of pretending cn8 covers them.
Every connection says what data moves and what code sits between: poll the job, chunk the transcript, convert VTT to text.
Seven languages — curl, Python, JavaScript, Java, C#, Go, PHP. Each API call is generated from the catalog; async services arrive with their polling loop already written.
Every generated architecture is stored under your profile and can be listed, re-opened, shared publicly or deleted. Re-downloading a language you already bought is free.
Scoping before you commit
See the whole pipeline, the prerequisites and the per-service pricing before writing a line of code.
First integration
Start from a project that already calls the right endpoints with the right fields, instead of stitching examples from the reference.
Sharing a design
Make a blueprint public and send the link to the person who has to approve or build it.
input
A conversation. Plain language, any language.
output
A sketch, a blueprint (nodes, edges, external boxes, prerequisites, notes, pricing) or a code project (files + run steps).
modestring · default: sketchsketch answers conversationally and is free (0 units). blueprint returns the architecture (1 unit). code returns a runnable project for one language (1 unit).
enum: sketch, blueprint, code
messagesrequiredarrayThe conversation so far: [{role: user|assistant, content: string}]. Write in any language; the answer comes back in the same one.
plan_idstringcode mode: the saved blueprint to implement (data.plan.id from a blueprint run).
planobjectcode mode: the blueprint inline, when you do not want to use a saved one.
languagestringcode mode: the one language to write. Omit it and the call is free: the answer comes back with needsLanguage:true and the list of options, so you can ask the user instead of paying for a language they did not want. One language per paid run; a language already generated for a saved blueprint is free to fetch again.
enum: curl, python, javascript, java, csharp, go, php
saveboolean · default: trueblueprint mode: store the result under your profile. Set false for a throwaway run.
modelstringOverride the planning model. Only allow-listed models are accepted; anything else falls back to the default.
50 credits for a blueprint, 50 credits for a code project. Conversation is free.
mode=sketch always reports 0 units — talk as long as you like before deciding.
A failed run (the planner could not map the idea to products, or the model did not answer) reports 0 units and is not billed.
Asking for a language you already generated on a saved blueprint is served from storage and bills 0.