#agentweb #extraction #sdf #structured-data

PageIQ

Structured web understanding for AI agents — powered by SDF Protocol

Install

openclaw plugins install clawhub:pageiq

AgentWeb

Structured web understanding for AI agents. Powered by SDF Protocol.

73,000 tokens of raw HTML → 886 tokens of structured data. 98% reduction.

Install

bash

clawhub package install agentweb

Or from source:

bash

pip install -e .

How it works

text

Agent needs a webpage
  → Check SDF cache (instant, 886 tokens avg)
  → HIT: structured entities, claims, relationships
  → MISS: your agent extracts → uses immediately → submits for credits
  → Next agent gets a cache HIT

Your agent's own LLM is the extraction engine. No extra models, no extra API calls.

Quick start

python

from agentweb_engine import AgentWebEngine

engine = AgentWebEngine(
    api_key="sdf_your_key",
    llm_fn=lambda prompt: my_llm.generate(prompt),
    model_provider="ollama",
    model_name="qwen3.5:9b",
)

result = engine.extract("https://example.com/pricing")
print(result.sdf["summary"]["one_line"])
# → "Example.com offers three pricing tiers: Free, Pro ($29/mo), Enterprise (custom)"
print(result.token_count)
# → 742 (vs ~65,000 from raw HTML)

Credit system

Every cache miss you fill earns credits:

Action	Credits
Submit extraction (gold quality)	+5
Submit extraction (silver quality)	+3
New URL (not in index)	+2 bonus
Cache lookup (hit)	-1

Check your balance:

python

balance = engine.get_credits()
print(f"Available: {balance.available}, Pending: {balance.pending}")

License

MIT — SDF Protocol