brand-loom

Open-source marketing skills that run on any model — with a hosted brand-memory engine. No coding agent required.

brand-loom is a small library of commodity marketing skills — hooks, captions, hashtags, content repurposing, SEO outlines, FAQs, schema.org markup, CTAs — that run on any LLM (OpenAI, Anthropic, Gemini, or a local model via Ollama). Bring your own key. No coding agent, no vendor lock-in.

Want it on-brand and multi-platform automatically — without writing prompts? That's the hosted engine at neoxra.com (auto voice extraction, multi-platform orchestration, Brand Kit). brand-loom is the open core; Neoxra hosts the rest.

Why brand-loom

• Model-agnostic. One interface, four providers, BYOK. Swap models with one env var.
• No coding agent required. Plain Python — not coupled to Claude Code / Cursor / OpenClaw.
• Brand-aware seam. Every skill accepts an optional brand_context. Ship a basic default here; get automatic, extracted brand voice on hosted Neoxra.

Quickstart (60 seconds, no API key)

pip install brand-loom

from brand_loom.providers import use_provider
from brand_loom.skills.hook import HookSkill

use_provider("fake")              # runs with zero keys; swap for "openai"/"anthropic"/"gemini"/"ollama"
skill = HookSkill()
print(skill.run_topic("How we cut cloud costs 40%"))

Real provider (BYOK):

pip install "brand-loom[openai]"
export OPENAI_API_KEY=sk-...
export BRANDLOOM_PROVIDER=openai

Run it as an agent (no coding agent required):

brand-loom run hook --text "How we cut cloud costs 40%"
brand-loom chain hook,caption --text "How we cut cloud costs 40%" --brand brand.json

A thin, linear runner — one skill or a simple chain, on any model. Sophisticated multi-platform orchestration (voice-matched, Planner→Critic, auto fan-out) lives in hosted Neoxra.

Agent chain (Python API):

from brand_loom.agent import run_chain, run_skill
from brand_loom.providers import use_provider

use_provider("fake")  # or "openai", "anthropic", "gemini", "ollama"

# Single skill
result = run_skill("hook", "How we cut cloud costs 40%")
print(result.text)

# Chain: hook → caption
result = run_chain(["hook", "caption"], "How we cut cloud costs 40%")
print(result.text)

# With brand context
result = run_skill("hook", "AI marketing", brand_context={
    "tone": ["bold", "witty"],
    "audience": "startup founders",
})
print(result.text)

Install in your coding agent

brand-loom also works as an agent skill in Claude Code, Cursor, OpenClaw, and any skills.sh-compatible host:

npx skills add hogan-tech/brand-loom

Each skill is available as a SKILL.md in the skills/ directory with frontmatter that agent hosts parse automatically. No configuration needed — install and use.

brand-loom remains model-agnostic and usable without any coding agent via pip install brand-loom + the CLI. The agent-skill surface is additive distribution, not a requirement.

Skills (v0.1)

Providers

The brand_context seam

Skills accept an optional brand_context dict (tone, audience, do_phrases, avoid_phrases). brand-loom ships an empty/basic default. Automatic voice extraction from your existing content, multi-platform fan-out, and a no-code UI live in hosted Neoxra — see HOSTED.md.

Contributing

PRs welcome — read CONTRIBUTING.md, SCOPE.md, and sign the CLA. In scope: single-shot commodity skills. Out of scope: voice engine, orchestration, tuned scorers, analytics/publish integrations (these live in the hosted product).

License

Apache-2.0. Hosted by Neoxra · a Meridian Global product.