For Developers

Drop an expert's brain into your agent. On demand.

Two types of skillbooks — references for facts and guides for method. Swap them into your agent over plain HTTP and it gains deep domain expertise instantly. No SDK. No retraining. No RAG.

Start Using Skillbooks → See How It Works

Just HTTP

Cited sources

Micropayments baked in

Data only, no code exec

The Challenge

Your agent is running on junk food. All three options are broken.

Training Data

Knowledge gets baked in — then goes stale the moment training ends. Regulations change, standards update, best practices evolve. Your agent can't tell you what it doesn't know, and it doesn't know what it doesn't know.

Web Search

Results optimize for clicks, not accuracy. Agents need signal, not SEO noise. Extracting reliable answers from web pages means parsing ads, paywalls, and content designed for humans browsing — not machines reasoning.

RAG Pipelines

Chunk boundaries split context. Embeddings drift. Your team builds and maintains the vector database, embedding pipeline, and retrieval logic — forever. That's a lot of infrastructure just to approximate what structured knowledge ships for free.

How It Works

Reference + Guide = Capability

Fetch SKILL.md

The free entrypoint. No payment, no auth. Your agent reads the type (reference or guide), scope, and table of contents.

Navigate the Graph

Follow links to relevant sections. The graph structure means your agent takes the shortest path to the right answer instead of scanning everything.

Pay Per Page

Micropayments via API key. Your agent fetches only the pages it needs. No bulk purchases, no subscriptions. Pay for what you use.

Cite the Source

Every page includes source metadata—regulation numbers, section references, publication dates. Your agent can cite exactly where the answer came from.

In Practice

The HTTP Flow

A complete interaction from discovery to cited answer. No SDK, no client library—just HTTP.

Terminal

# 1. Fetch a reference (source material)
curl https://skillbooks.ai/eu-ai-act/SKILL.md

# Response: table of contents with links to every section
---
skillbook-type: reference
---
# EU AI Act — Compliance Guide
## Table of Contents
- [Article 6 — High-Risk Classification](./classification/article-6.md)
- [Annex III — High-Risk AI Systems](./classification/annex-iii.md)
- [Article 9 — Risk Management](./risk-management/article-9.md)
- [Article 52 — Transparency Obligations](./transparency/article-52.md)
...

# 2. Agent needs Article 6 classification requirements
curl -H "Authorization: Bearer sk_live_..." \
     https://skillbooks.ai/eu-ai-act/classification/article-6.md

# Response: structured content with citations
---
source: EU AI Act, Article 6, OJ L 2024/1689
effective: 2024-08-01
---
# Article 6 — Classification Rules for High-Risk AI Systems
AI systems referred to in Annex III shall be considered
high-risk and subject to the requirements of Chapter III...


# 3. Agent responds to the user with a cited answer
"AI systems listed in Annex III are classified as high-risk
under Article 6 and subject to Chapter III requirements."
— Source: EU AI Act Skillbook, Article 6, OJ L 2024/1689

Integration

Load expertise the way you install a package. It's just HTTP.

Any Framework

LangChain, CrewAI, OpenClaw, or your own custom agent loop. If it can make HTTP requests, it can read a skillbook. No vendor lock-in, ever.

Three Steps Total

Fetch SKILL.md. Navigate to the relevant page. Cite the source in your response. That's the entire integration. Seriously.

No Infra to Maintain

No vector database. No embedding pipeline. No chunking strategy debates. The structure ships with the content. You just fetch it and go.

Python

import requests

# Fetch a reference (facts) and a guide (method)
reference = requests.get("https://skillbooks.ai/eu-ai-act/SKILL.md").text
guide = requests.get("https://skillbooks.ai/compliance-audit/SKILL.md").text

# Agent reads both TOCs, composes reference + guide
ref_pages = agent.select_pages(reference, user_question)
guide_steps = agent.select_pages(guide, user_question)

# Fetch paid pages from both types
headers = {"Authorization": f"Bearer {API_KEY}"}
context = [requests.get(url, headers=headers).text
           for url in ref_pages + guide_steps]

# Agent composes: applies guide method to reference facts
answer = agent.compose(user_question, context=context, cite=True)

Comparison

Honest Tradeoffs

Every approach has strengths. Here is how skillbooks compare on the dimensions that matter for production agent systems.

	Skillbooks	RAG Pipeline	Fine-Tuning	Web Search
Time to first answer	Minutes	Days	Weeks	Minutes
Infra you maintain	None	Vector DB + pipeline	Training workflow	None
Source citations	Built-in	Possible	No	Unreliable
Stays current	Yes	Reindex	Retrain	Yes
Context quality	Structured pages	Chunk dependent	Embedded	Noisy
Best fit	Expert knowledge layer	Private corpora	Behavior shaping	Fresh discovery
Tradeoff you accept	Pay for premium knowledge	Own the ops burden	Slow iteration	Lower trust
Composable methods	Reference + Guide	No	No	No

Your Agent Is Ready. The Knowledge Is Waiting.

Load domain expertise into any agent, any framework, over plain HTTP. Nothing to install. No infra to maintain. Just results.

Get Started →