Quick Start

Get evidence recording working in under 3 minutes.

Install

Python

pip install atlast-ecp

TypeScript

npm install atlast-ecp-ts

Initialize

atlast init

This creates:

• ~/.ecp/ — local evidence storage (never transmitted)
• Ed25519 keypair for signing
• Agent DID (did:ecp:{hash})

Option A: Zero-Code (Proxy)

Wrap any existing agent — no code changes:

atlast run python my_agent.py

This starts a transparent proxy that intercepts all LLM API calls (OpenAI, Anthropic, etc.) and records evidence automatically.

Option B: SDK Wrapper (5 lines)

from openai import OpenAI
from atlast_ecp import wrap

client = wrap(OpenAI())  # That's it

# Use normally — every call is automatically recorded
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello"}]
)

Option C: Explicit Recording

from atlast_ecp import record

record(
    in_content="Summarize this document",
    out_content="The document discusses...",
    model="gpt-4",
    latency_ms=1200,
)

View Your Evidence

# See recent records
atlast log

# Verify a specific record
atlast verify rec_abc123

# View stats
atlast stats

# Inspect with original content
atlast inspect rec_abc123

Upload to Server

# Register your agent
atlast register

# Push evidence batch
atlast push

This uploads Merkle roots to api.weba0.com for on-chain anchoring.

Super-Batch Aggregation

When 5+ batches are pending, the server automatically aggregates them into a single on-chain attestation via a Super Merkle Tree — reducing gas costs by up to 1000×.

Each batch gets an inclusion proof that independently verifies it belongs to the on-chain attestation, without requiring the other batches.

┌─────────────────────────────────────┐
│         Super Merkle Root           │ ← 1 EAS attestation
│      ┌────────┴────────┐            │
│   Hash(A+B)       Hash(C+D)        │
│   ┌──┴──┐        ┌──┴──┐           │
│ Root_A Root_B  Root_C Root_D        │ ← Individual batch roots
└─────────────────────────────────────┘

You don't need to do anything — super-batching is automatic. Your batch's inclusion_proof is included in the webhook payload for independent verification.

Query a super-batch:

curl https://api.weba0.com/v1/super-batches/{super_batch_id}

Behavioral Drift Detection

The server monitors your agent's behavior patterns over time. If record counts, latency, or other metrics shift significantly, drift is detected:

curl https://api.weba0.com/v1/agents/{did}/drift

Returns:

{
  "drift_score": 0.12,
  "drift_detected": false,
  "baseline_window": 20,
  "current_window": 5
}

Next Steps

• Core Concepts — understand ECP data model
• Python SDK — detailed SDK reference
• TypeScript SDK — Node.js integration
• API Reference — server endpoints