Why AI Agents Need WebWeaveX
Compact IR Graphs
500KB HTML blobs become 15KB structured IR graphs for LLM context windows.
Session Continuity
Persists authenticated state across tool calls without re-authenticating.
Deterministic Hashes
SHA-256 state digests verify action success with mathematical certainty.
Replay Proofs
Verify topological equivalence between before/after states.
Sandbox Execution
Only allowlisted transitions are executed. No arbitrary code eval.
Cross-Language
Same output whether calling Python, JS, Dart, Java, or Kotlin.
Universal Agent Workflow Cycle
flowchart TD
A[1. Ingest & Extract] --> B[2. Cognize IR Graph]
B --> C[3. Execute Action]
C --> D[4. Verify Hash]
D -->|Match| E[5. Advance Tick]
D -->|Mismatch| F[6. Alert & Rollback]
Recommended System Prompt
system_prompt.md
You are an autonomous operational software agent powered by WebWeaveX.
When interacting with target applications:
1. Always parse inputs via `run_canonical_pipeline(UniversalInput(...))`.
2. Inspect `graph.nodes` and `fingerprint` to locate interactive elements.
3. Pass `encrypted_session` in subsequent calls to maintain state.
4. Compare `pipeline_hash` before and after execution to verify transitions.
5. Only execute allowlisted state transitions.
6. If hash diverges, halt and report the anomaly.LangChain Integration Recipe
webweavex_tool.py
from langchain.tools import tool
from webweavex import UniversalInput, run_canonical_pipeline
@tool
def inspect_runtime_surface(url: str, session_token: str = None) -> dict:
"""Inspect a web app's runtime surface."""
input_data = UniversalInput(
source=url,
source_type="web",
session={"auth_token": session_token} if session_token else None
)
result = run_canonical_pipeline(input_data)
return {
"pipeline_hash": result.pipeline_hash,
"node_count": len(result.graph.nodes),
"nodes": [n.node_id for n in result.graph.nodes[:20]],
"encrypted_session": result.encrypted_session
}Tool Calling Pattern
agent_loop.py
from webweavex import UniversalInput, run_canonical_pipeline
# Initial extraction
initial = run_canonical_pipeline(UniversalInput(
source="https://app.example.com",
source_type="web",
session={"auth_token": "user_token_abc"}
))
baseline_hash = initial.pipeline_hash
# Agent performs actions...
# Re-extract and verify
after = run_canonical_pipeline(UniversalInput(
source="https://app.example.com",
source_type="web",
session={"auth_token": "user_token_abc"},
previous_session=initial.encrypted_session
))
if after.pipeline_hash != baseline_hash:
print(f"State changed: {baseline_hash[:16]} -> {after.pipeline_hash[:16]}")