Canvora × nanoclaw

On-brand visuals CLI for nanoclaw

nanoclaw keeps the agent loop minimal — Canvora keeps the tool surface minimal to match: one env var, one command in, finished branded visuals out.

Start free Setup guide

Free credits, no card required. Please make sure there is always a human in the loop.

The whole setup

Terminal

# Install the skills (or use the CLI bare - both work)
npx skills add canvora/canvora-mcp

# Auth is one env var
export CANVORA_API_KEY=vd_...

# Smoke test
canvora credits --json

This is what your agent ships

One generation, localized twice. Same design, natively re-composed in Spanish and Arabic — right-to-left done correctly, not mirrored English.

Generated with the same commands shown on this page.

How it works

1. Fire

First stdout line is the bare generation id

Terminal

ID=$(canvora generate \
  --input "Three habits of highly effective remote teams..." \
  --format instagram_post --style minimal | head -1)

2. Poll

Cheap status checks until completed

Terminal

while true; do
  S=$(canvora status "$ID" --json | jq -r .status)
  [ "$S" = "completed" ] && break
  [ "$S" = "failed" ] && exit 1
  sleep 10
done

3. Deliver

Files or CDN URLs, whichever the loop needs

Terminal

canvora download "$ID" --dir out/
canvora status "$ID" --json | jq -r '.outputs[].fileUrl'

What is nanoclaw?

nanoclaw is the minimal take on the personal AI agent: a tiny, auditable core loop that leans on well-behaved external tools instead of built-in complexity. Canvora fits that philosophy — the entire visual pipeline is four commands.

Tiny surface

generate, status, download — the essential loop is three commands. Everything prints JSON with --json; progress goes to stderr, results to stdout.

No state to manage

No sessions, no SDK, no config files. The API key comes from the environment; generation ids are plain strings your loop passes around.

Fails safe

Balance is checked before any deduction, errors are typed JSON, and failed generations refund automatically — a minimal loop can't silently drain anything.

The entire agent tool, one screen

This is the whole integration — no wrapper library

Terminal

#!/bin/sh
# nanoclaw tool: text -> branded instagram post -> file URL
ID=$(canvora generate --input "$1" \
  --format instagram_post --brand "$BRAND_ID" | head -1)
until [ "$(canvora status "$ID" --json | jq -r .status)" = "completed" ]
do sleep 10; done
canvora status "$ID" --json | jq -r '.outputs[0].fileUrl'

Add languages without adding code

The same minimal loop covers every market

Terminal

# One extra command localizes the finished visual -
# same design, natively re-composed text:
canvora localize "$ID" --language tr
canvora localize "$ID" --language ar   # correct RTL, not mirrored

Why nanoclaw + Canvora

Three-command core loop: generate, status, download
stdout is data, stderr is progress — pipes compose cleanly
No SDK, no session, no config: one env var and plain strings
Typed JSON errors and pre-deduction balance checks — safe in unattended loops
localize and variations extend the loop without new integration code

Command reference

Add --json to any command for machine-readable output. Progress goes to stderr, results to stdout.

Command	Description
canvora generate	Create visuals from --idea, --input text, --url or --file-url; pick formats, brand, style, language
canvora status <id>	Check a generation; --json returns status and output file URLs
canvora download <id>	Save output files to a directory
canvora localize <id>	Same design, natively re-composed in another language (RTL included)
canvora variations <id>	Alternate takes on a finished generation
canvora edit <outputId>	Revise one visual with a natural-language instruction
canvora formats	List all 100+ format ids
canvora brands	List brand kits (colors, fonts, logo, tone)
canvora credits	Show credit balance

100+ formats, one engine

Every format has platform-correct composition and safe zones — your agent picks by id.

Instagram Carousel

Up to 20 slides, hook to CTA

LinkedIn Carousel

Swipeable document posts

Pitch Deck

Coherent multi-slide sets

Instagram Post & Story

Feed and vertical formats

Stat Card

One number, designed to land

Infographic

Data and process visuals

Blog Header & OG Image

The launch surface

Ad Creatives

Display and social ads

See all formats and tools →

FAQ

Why is generation asynchronous? +

Visual generation is real work; a blocking call would tie up the agent loop and time out shells. Fire the job, capture the id from stdout, poll status. The pattern is three lines of sh, shown above, and it never stalls.

What does each call cost? +

Flat credits: 10 per single visual, 15 per carousel or deck slide. canvora credits --json returns the balance; a request that exceeds it fails before any deduction with the exact shortfall in the error.

What if the loop dies mid-generation? +

Nothing is lost: the generation continues server-side and the id remains queryable. If it fails, credits refund automatically. Re-running the script starts a new, separate job — so persist the id if you want to resume instead.

Is there anything to install besides the CLI? +

No. @canvora/cli is a plain Node binary; jq in the examples is optional sugar. The skills install adds ready-made workflows on top, but the bare CLI is fully capable.

Also works with

Hermes Agent → Codex → Any MCP client →

Your agent's next tool call is a design department.

Create a free account, set one key, and let nanoclaw ship on-brand visuals.

Start free Read the setup guide